Physical AI Brief

To effectively integrate AI into high-stakes, critical environments such as healthcare, autonomous driving, and aviation--and to advance toward higher levels of automation and seamless human-AI collaboration--building trust in AI-driven solutions is essential. Trust, in turn, is closely linked to the explainability of AI systems. The rapid advancements in AI across various domains have underscored the challenges of establishing trust, raising increasing interest in AI explainability even more when applied to deep learning. In this context, the present work aims to explore the application of explainability techniques to Reinforcement Learning (RL) algorithms, specifically within the safety-critical domain of Air Traffic Control (ATC). Using a simplified ATC environment as an initial testbed, an intelligent agent is trained with a reinforcement learning algorithm to make decisions on alternative flight routes that avoid no-fly zones. As a preliminary explainability approach, a saliency map is employed, providing insights into the input features that most significantly influence the agent's decision-making process.

Adding procedural skills to an LLM agent is typically evaluated by average improvement in task success. However, this metric hides an important cost: skills can also make agents worse. We measure both sides by comparing agents with and without skills across nearly 6,000 runs spanning two office automation benchmarks and three model harness stacks. This allows us to distinguish two outcomes. A regression is a task solved without skills but failed after skills are added. A residual failure is a task that fails both with and without skills. We find that regressions are substantial enough that the best performing skills outperform others primarily by regressing less, not by gaining more. We identify three causes of regression: (i) skill description osmosis, a skill changes an agent's behavior simply by being present in context, even when it is never invoked; (ii) grounding displacement, a skill's prescribed procedure overrides how the agent interprets its inputs; and (iii) verification displacement, where the procedure suppresses checks the agent would otherwise perform on its outputs. Analysing persistent failures reveals the same underlying pattern. Existing skills overemphasize procedural guidance the stage least often responsible for failure while under supporting grounding and verification, the dominant sources of remaining errors. After correcting evaluation artifacts and studying traces, we find many regressions and persistent failures recoverable through better grounding and verification. Procedural skills should be evaluated by decomposing their net effect into gains and regressions, not by aggregate improvement alone. We identify three regression modes skills should avoid, and find that reliability depends more on grounding and verification than on procedural skill choice.

A central design principle in modern machine learning and artificial intelligence is to align a model's inductive bias with the structure of its input data. For matrix-valued inputs, relevant matrix-level relationships can be characterised through spectral values and spectral subspaces; however, common coordinate-wise rotation-gate data-encoding unitaries used in most quantum machine learning models do not explicitly construct such a matrix-level representation. We introduce Quantum Spectral Models (QSMs), in which we construct the generator of the data-encoding unitary directly from each input matrix. We study three QSM variants based on symmetric, global block, and non-overlapping patch-local block Hamiltonians. Their outputs admit truncated Fourier representations in which input-dependent spectral gaps supply candidate phase carriers, while spectral subspaces help determine their coefficients. We evaluate the QSMs and comparison quantum models on two matrix representations of Pendigits and two controlled synthetic tasks defined by spectral statistics. At the largest evaluated circuit depth, QSM variants lead the tested quantum models in mean test accuracy across all four benchmarks. The patch-local QSM leads on Pendigits, whereas the global block-Hamiltonian QSM leads on the controlled spectral tasks. Ablations show a task-dependent reversal: subspace-preserving controls perform better on Pendigits, whereas spectral-value-only controls lead among the tested ablations on the synthetic tasks. Together, these results shed new light on quantum machine-learning model design by showing how input-conditioned spectral representations can provide an analysable inductive bias, while offering a broader perspective on structure-aware model design in machine learning and artificial intelligence.

Automating theoretical research is constrained not only by the generation of candidate results, but also by their reliable evaluation. A common approach is to close the research loop with a large language model (LLM) reviewer. However, such reviewers remain empirically unreliable: they may accept fabricated papers and detect them at rates close to chance (Bad Scientist, 2025). We present CausalForge, a framework for automated theoretical research in causal inference grounded in the Lean proof assistant. CausalForge combines Causalean, a foundational Lean library for causal inference containing 7,035 machine-checked declarations developed with language-model assistance under human design and review, with CausalSmith, a self-improving agentic pipeline that selects research topics, proposes results, formalizes statements, constructs proofs, and presents the resulting artifacts for human inspection. Because a machine-checked proof establishes only that a formal statement follows from its assumptions, not that the statement faithfully captures the intended scientific claim, the pipeline augments kernel verification with a statement audit that compares each formal theorem against the informal claim it is intended to express. We evaluate the system using artifacts produced by completed autonomous research runs. The source code, formal library, and run records are available at https://github.com/Jiyuan-Tan/CausalForge.

Electroencephalography (EEG) is widely used to diagnose neurological conditions, but its analysis usually relies on either predefined spectral features or deep neural networks. Predefined features carry a strong bias, since they fix in advance what counts as informative, while deep neural networks and foundation models are hard to interpret and need large amounts of data and compute. We present bag-of-waves, an interpretable framework that learns a small dictionary of recurring EEG waveform templates, called atoms, using shift-invariant k-means without labels. The continuous EEG is then turned into a sequence of atom tokens, whose counts feed a simple downstream classifier or clustering step. We extend this representation in two ways: we add atom-to-atom transitions, which we call n- grams, to capture temporal structure, and we move from single-channel atoms to regional and cross-channel spatial atoms for the multichannel case. We test the method on three complementary datasets, each probing a different aspect: single-channel mouse genotype clustering with only sixteen animals (the low-data and temporal case), resting-state dementia classification (the spatial case), and the TUEV benchmark, a six-way classification of clinical EEG events (a high-data comparison against strong deep and foundation baselines). Across all three datasets, bag-of-waves achieves performance competitive with state-of-the-art deep and foundation models. Yet, it operates with a fraction of the parameter count and provides full interpretability: because every atom corresponds to an inspectable waveform, the method explicitly recovers known clinical morphologies that a neurophysiologist can directly validate. Its main advantage is that it works in the low-data regime where heavier models are a poor fit.

Low-Rank Adaptation (LoRA) has become a widely adopted technique for efficient neural network fine-tuning, decomposing model updates into low-rank matrices. However, LoRA remains computationally costly because it updates all matrices uniformly, regardless of their actual contribution to adaptation. This cost is especially prohibitive for large-scale models with billions of parameters and for resource-constrained settings such as edge deployment and on-device fine-tuning. We show for the first time that not all LoRA matrices are equally worth tuning: matrices with smaller condition numbers (the ratio of largest to smallest singular value) are already well-balanced across directions and contribute only marginally to adaptation, whereas matrices with larger condition numbers contain underdeveloped directions that span richer subspaces and drive most of the performance gains. This observation itself is a key contribution of our work, and it motivates a more selective approach to fine-tuning. Building on this insight, we propose \k{appa}-LoRA, a method that optimizes LoRA by focusing updates on the matrices with the largest condition numbers, which capture the most informative directions of change. By restricting LoRA updates to the top 50% of weight matrices ranked by condition number, \k{appa}-LoRA halves the trainable parameter count and correspondingly reduces compute and memory cost. Extensive experiments across multiple benchmarks show that this design cuts fine-tuning time by 16.2% on average while matching the accuracy of standard LoRA and reducing memory cost by 4.5%. Further analysis reveals that the condition numbers of the selected matrices consistently decrease over training, suggesting that \k{appa}-LoRA's effectiveness stems from targeted spectral rebalancing rather than parameter selection alone.

Large Language Model (LLM)-based Test-Driven Development (TDD) has advanced automated code generation. However, existing approaches depend heavily on human-crafted test cases and cannot operate effectively when only natural-language requirements are available. Although recent work enables automatic test generation, it often overlooks the inherent stochasticity of LLMs, leading to two key defects: faulty tests generate misleading feedback that distorts code optimization, while mixed-quality test cases produce conflicting evaluation signals that hinder reliable code selection. To address these challenges, we propose MineValiCoder, a collaborative closed-loop TDD framework based on the mutual reinforcement of test-case quality and code quality. MineValiCoder comprises three modules. The Test Case Quality Mining (TCQM) module filters faulty test cases through self-validation, providing reliable optimization supervision. The Parallel TDD Refinement module iteratively optimizes code and generates diverse high-quality code candidates using validated test-case feedback. The Bipartite Graph-Based Code-Test Mutual Validation (BiCoTeV) module dynamically models code-test interactions and performs mutual validation scoring for stable and reliable optimal-code selection. Extensive evaluations across four LLMs and mainstream benchmarks show that MineValiCoder significantly outperforms state-of-the-art methods. Specifically, it achieves Pass@1 scores of 96.34% on HumanEval, 87.40% on MBPP, 64.00% on APPS, and 51.33% on LiveCodeBench. These results demonstrate the effectiveness of MineValiCoder in mitigating LLM stochasticity and improving the reliability of automated code generation.

Routing to select large language models (LLMs) with different cost-quality trade-offs has become a fundamental deployment feature of enterprise AI. Existing routers, primarily make independent routing decisions for each LLM call. However, agentic applications execute as long-horizon workflows whose quality is determined only by a delayed, task-level outcome. This mismatch prevents per-call routers from correctly attributing feedback to individual routing decisions. Towards mitigating this, we present TRACE-Router, a task-level routing framework that aligns routing with the unit of supervision. TRACE-Router assigns each task to a model once at admission using a contextual bandit, pins all subsequent LLM calls to the selected backend, and updates its policy using the task's terminal reward, jointly accounting for accuracy and latency. By leveraging delayed task feedback, TRACE-Router learns routing policies that adapt to the workload while avoiding explicit task-complexity estimation. Across three agentic benchmarks, TRACE-Router consistently improves the accuracy-latency trade-off, achieving non-dominated Pareto frontier points. On tau2-Bench, it outperforms latency-matched interpolation between individual models by 7-8 accuracy points, while on Terminal-Bench it achieves 7.1 higher accuracy points than the strongest single model baseline with 36% lower latency.

Enterprise AI agents are typically granted static credential sets at configuration time, holding every tool the role might need for every task they perform. This persistent over-privilege expands the attack surface. We argue that capability scoping must follow a dynamic least-privilege principle and be treated as a prevention mechanism before a detection one. A credential that does not exist in an agent's context cannot be misused regardless of the agent's reasoning or evasion sophistication. We outline a three-source architecture instantiating this principle: role-based ceilings, a task-context classifier, and policy-derived combination prohibitions creating a layered proactive defense against LLM agent misalignment and misuse cases. The architecture supports both enforcing and observe-only deployment; the latter records agent permission requests inconsistent with task context, producing a behavioral signal usable in misalignment research. As a first step toward evaluating this architecture, we contribute a synthetic dataset of 600 enterprise task prompts grounded in a multi-department company policy, labeled with minimum required permissions across a 15-permission tool-based taxonomy that maps directly to deployable credentials or enforceable guardrails. The dataset is constructed via a two-pass pipeline that separates prompt generation from permission labeling to avoid circularity, and is validated against a 60-record/688 decisions human-reviewed sample (Cohen's $κ= 0.917$ pre-review and $κ= 0.967$ post-review). Iterating between dataset and policy reduced ceiling violations from 46 to 3, a 93% reduction. This shows that synthetic prompt generation can drive policy refinement when the two are developed together. The dataset, environment specification, and generation pipeline are released to support evaluation of dynamic scoping mechanisms.

This work addresses the generation of theoretical correlation matrices with prescribed sparsity patterns associated to graph structures. We propose a novel convex optimization framework in which an initial matrix is projected onto an elliptope under a positive semidefiniteness constraint. Several numerical schemes are implemented and compared. The problem falls within the broader class of matrix completion, where off-diagonal entries corresponding to absent edges are fixed to zero and diagonal entries are fixed to one. Beyond this structural constraint, the approach offers greater flexibility than existing methods by allowing control over the mean of the off-diagonal entry distribution, enabling the generation of correlation matrices that better reflect realistic data. This procedure is not designed to yield a uniform distribution over the feasible set; rather, it provides a principled and tunable way to construct correlation matrices suitable for benchmarking statistical methods for graphical model inference. Theoretical guarantees on the existence of solutions are established, both in the general setting and under the additional mean constraint. Simulation studies illustrate the properties of the generated matrices with respect to graph structure. The methodology is applied to two real-world datasets from neuroscience and finance, and a comparison with GAN-based correlation matrix generation is provided.

Humans routinely communicate through abstractions of their bodies, including shadows, silhouettes, and reflections. Yet robots remain largely confined to expressing themselves through their physical morphology. Enabling robots to communicate through such projected visual abstractions requires reasoning not only about bodily motion but also about how that motion is transformed into an external representation perceived by an observer. Among these abstractions, shadows provide a particularly compelling example because they emerge directly from the robot's embodiment while remaining visually distinct from the body itself. Here, we present a robotic system capable of dynamic shadow expression using a 21-degree-of-freedom dexterous hand with compliant soft skin and a learned shadow self-model. The soft-skinned embodiment reduces light leakage to produce visually continuous silhouettes, while the differentiable self-model learns the mapping between hand configurations and projected shadow appearance through task-agnostic self-exploration. Given a target shadow image or video, the robot optimizes its hand configurations through gradient-based search over 1 the learned self-model and refines the solution through collision-aware simulation to obtain physically feasible motions. For dynamic shadow performance, we further introduce expressive-region objectives, temporal smoothness regularization, and keyframe-based optimization to preserve visually important motion cues while reducing optimization complexity. We demonstrate robotic shadow expression across sign-language gestures, hand-shadow puppetry, and animal motion imitation in both simulation and physical experiments. These results establish a framework for enabling robots to manipulate projected visual abstractions of themselves for communication and visual storytelling.

Learning world models that infer environment dynamics from high-dimensional observations and predict outcomes under candidate actions is central to planning and control. Joint-Embedding Predictive Architectures (JEPAs) provide a compelling framework for learning such models in representation space. Recent action-conditioned extensions perform promisingly in visual control and latent-space planning, but leave a fundamental question unresolved: when does controlled latent prediction identify both the underlying state and the controlled dynamics? This is challenging under nonlinear observations and behavior policies with limited conditional action variation, where state-dependent evolution and action effects can be statistically confounded. We establish a joint identifiability theory for controlled world models with Gaussian latent states under state-dependent Gaussian behavior policies. We identify two policy-dependent conditions: spectral separation of the predictable signal governs representation identifiability, while non-degenerate conditional action variation governs transition identifiability. We prove that when both conditions hold, every global minimizer of the JEPA objective identifies the latent state and controlled transition up to an orthogonal transformation. We further derive quantitative bounds on representation and transition identifiability under approximate optimization. Finally, we construct predictor perturbations along weakly excited action directions whose counterfactual-to-on-policy error ratio is the inverse transition-identifiability margin, revealing the cost of limited action coverage. Experiments across nonlinear observation maps and behavior policies corroborate the theory and demonstrate implications for transition identifiability, counterfactual prediction, and goal-conditioned latent planning.

Accurate on-device fluid identification is essential for microfluidic applications, yet maintaining reliability under varying flow, pressure, and temperature remains a key challenge. Existing learning-based methods often treat sensor signals as domain-agnostic features, neglecting the underlying physical relationships that govern fluid behavior, thereby limiting generalization and interpretability. To address this, we propose PRIMS, a physics-aware multimodal Transformer that integrates physical knowledge into representation learning and attention mechanisms through three dedicated modules: (1) Physics-based Token Vectorization transforms raw Coriolis and pressure sensor signals into physically meaningful token embeddings; (2) Physical Component Synthesizer models viscosity-related dependencies among flow, pressure, and density; and (3) Physics-guided Fusion captures cross-physical correlations through attention-based integration. By embedding these physics-based relationships directly into the model architecture, PRIMS bridges analytical fluid mechanics and deep learning, enabling interpretable, data-efficient, and resilient fluid classification. Evaluations on a five-fluid benchmark under dynamic flow, pressure, and temperature conditions show that PRIMS achieves 98.92% average F1-score with only 0.46 million parameters, a 14 times reduction compared to state-of-the-art Transformer-based methods. PRIMS also consistently outperforms prior SOTA models under out-of-distribution shifts to unseen temperature ranges and unseen flow-rate ranges, indicating strong robustness to operating conditions not observed during training. These findings suggest that designing architectures that explicitly mirror governing physical relationships can make them learn transferable, environment-independent representations, improving real-world reliability for microfluidic sensing.

Large Language Models (LLMs) are increasingly deployed as autonomous agents, transitioning from static conversational interfaces to dynamic systems capable of complex reasoning, tool execution, and decision-making. However, the operational reliability of these agentic AI systems is fundamentally challenged by the absence of reliable ground truth in open-ended environments and the risk of increasing operational drift over time. To address this challenge, we propose and experimentally evaluate an agentic AI framework, designed to enforce autonomous integrity within LLM-driven systems. We design a self-calibration mechanism that mitigates drift and dynamically approximates ground truth by incorporating an ARIMA forecaster, without requiring continuous human oversight. To demonstrate the effectiveness and reliability of our methodology, we apply it to the complex domain of profiling the resource usage of zero-knowledge workloads in edge computing networks. Experimental results show that the proposed self-calibrating agentic framework successfully profiles the zero-knowledge workloads, achieving a higher accuracy than baseline LLM agents by 91.7% for resource usage prediction and improving the prediction speed by 71.7% compared to pure profiling, establishing a robust foundation for deploying autonomous AI in decentralized infrastructures. Furthermore, the ground truth generation using the proposed ARIMA leaping algorithm is 52% faster than a standard ARIMA forecasting algorithm, while achieving the same accuracy.

Vision-language model (VLM) agents increasingly use tools to act on 3D scenes rather than only describe them. Existing 3D benchmarks score textual responses or single-object operations, leaving agent action on complete multi-object 3D scenes under evaluated. We present SceneActBench, a benchmark for visually conditioned action across five 3D tasks under a unified agent-environment loop. Given PNG images or sampled video frames and, where applicable, supplied 3D assets, an agent acts on a 3D environment. We evaluate each final output against hidden ground truth with task-specific geometric metrics. SceneActBench comprises five tasks built from 210 source instances, yielding 520 task cases including paired input conditions. Every task runs through one fixed agent loop to keep the comparison fair. Across eleven proprietary VLM configurations, Overall scores span 38.6-50.2, and none performs consistently well across tasks. We further analyse where and how failures manifest.

With the rapid adoption of long-context large language models (LLMs), the continuously growing KV cache during decoding has become the critical memory bottleneck. To tackle this challenge, we propose HiKV, a novel algorithm-hardware co-design that exploits KV cache redundancy through hierarchical importance awareness. Algorithmically, HiKV compresses the KV cache at two granularities: Stage I evicts unimportant tokens within a fixed budget, and Stage II further loads only the significant elements of each retained token, reaching compression ratios unattainable at a single granularity. Architecturally, we develop a dedicated accelerator centered on a reconfigurable importance sorter that switches between the distinct sorting datapaths each stage requires, unifying the two-stage acceleration in one circuit with minimal overhead. Evaluated on representative LLMs, HiKV achieves up to 7.95x speedup and 90% energy reduction in the attention computation over the vanilla KV cache baseline within negligible 1% accuracy loss. Under iso-accuracy constraints, HiKV outperforms state-of-the-art importance-based methods by achieving an additional 1.82~4.87x reduction in external memory accesses. These benefits are enabled by specialized hardware components that add only 8% to the system area.

Diagnosing the root cause of anomalies is essential for safe industrial operation. Despite extensive sensor instrumentation, formulating hypotheses and gathering evidence remains a manual process, creating a major operational bottleneck. While existing data-driven approaches aim to automate this, two critical limitations restrict their deployment: their operate as black boxes unable to justify their diagnosis, and they require scarce labeled examples of faulty operation. To address this gap, we introduce AgentRCA, a zero-shot agentic framework for evidence-grounded root cause analysis. Rather than learning fault-specific mappings, AgentRCA performs inference-time reasoning by combining a data-driven digital twin (modeling normal system dynamics) with a tool-augmented large language model. The agent iteratively gathers statistical evidence, evaluates competing hypotheses, and identifies the physical fault that best explains the observed behavior. Evaluated on a real-world multiphase-flow facility and a large-scale chemical plant, AgentRCA achieves diagnostic performance competitive with fully supervised baselines without relying on fault-specific training. Crucially, it produces transparent reasoning traces that explicitly link observed symptoms to their underlying physical causes. These results establish autonomous hypothesis-driven reasoning as a practical foundation for scalable industrial root cause analysis.

Curvature notions on graphs, particularly Ollivier-Ricci and Forman, have emerged as powerful tools for addressing fundamental issues in Graph Neural Networks (GNNs) such as oversmoothing and oversquashing, but rely almost exclusively on local edge-level comparisons and therefore fail to certify how information actually propagates over long distances. We introduce Entropic Curvature, a global, transport-based curvature obtained by extending the Lott-Sturm-Villani framework to graphs through the displacement convexity of entropy along Wasserstein geodesics. We define a tractable Weak Entropic Curvature proxy that lower-bounds the global entropic curvature, and from it derive (i) a Poincare-type inequality controlling oversmoothing, (ii) a transport-entropy generalization bound, and (iii) an expansion paradox proving that sparsity, strong spectral expansion, and positive entropic curvature cannot coexist in large graphs, unifying oversmoothing and oversquashing as opposite ends of a single curvature spectrum. We translate the theory into three practical mechanisms, the E-Gate aggregator, the ENT structural encoding, and Midpoint-Completion Rewiring (MCR), and benchmark them against SDRF, FoSR, BORF, LCP, and Graph Ricci Flow on six node-classification benchmarks, and graph-classification.

Large Language Models (LLMs) have significantly automated the process of scientific discovery over the past few years. However, existing systems share one core limitation: they generate and optimize ideas independently for either Quality or Diversity. This often leads to the generation of ideas in close proximity to one another or to a large set of trivial, unsound, or unclear concepts. In this work, we instead argue that research ideation should be treated as a conjunction of both objectives and framed as a Quality-Diversity (QD) search. In line with this perspective, we introduce IDEAgent, a multi-agent framework that manages the evolution of ideas through lineages. We jointly drive Quality using multi-objective feedback for dedicated repair and refinement, while Diversity is achieved through lightweight sequential memory and explicit comparison against completed ideas, their historical ancestors, and rejected proposals. To systematically evaluate this QD conjunction, we develop Yield, a joint metric that computes the largest set of mutually diverse ideas that satisfy a predetermined quality threshold. Finally, through evaluations across 32 topics spanning 8 domains of Computer Science, we show that IDEAgent outperforms the best baseline by 3.89x on Yield, while achieving non-zero Yield on 8x more topics. We further corroborate these findings through an analysis of quality improvements, showing that repair and refinement are crucial for building logical rigor and clarity while preserving non-obviousness. To encourage future research on QD-search-based ideation, we open-source IDEAgent at https://github.com/declare-lab/IDEAgent.

Agent benchmarks increasingly evaluate repository editing, web research, terminal use, and long-horizon interaction. Their scores support capability claims only when the evaluation protocol keeps the intended capability necessary for success. Recent reward-hacking benchmarks and system reports show that agents can instead recover public solutions, read evaluation artifacts, infer generator structure, manipulate feedback, or benefit from invalid scoring paths; existing responses do not provide a common procedure for attributing these shortcuts and quantifying their effect across benchmarks. We formulate protocol validity and introduce HackDetect, a post-hoc audit that identifies an exposure, determines how the agent used it, and assesses whether the resulting score is misleading. We quantify score inflation with the Mislead gap, defined as the exploit score minus the intended score. We audit 2,385 traces across 15 agent benchmarks and find evidence of exposures and reward hacking in 67.0% of Frontier Science traces and 66.7% of AutoLab tasks. Across paired comparisons, we measure score inflation of 0.45-1.00, showing that benchmark reports should provide evidence that scores reflect the intended capability.

High-complexity operational environments require methods that detect and anticipate temporally distributed patterns rather than classify isolated events. This paper introduces TRACTA (Temporal Reasoning and Capability-Trajectory Analysis), a controlled synthetic benchmark for temporal structural reasoning in high-complexity event-driven systems, instantiated through Multi-Domain Operations (MDO)-like scenarios. The benchmark includes three tasks: early_warning, pattern_detection, and run_classification, and compares raw-event neural models, a contract-lite semantic baseline, and a neuro-symbolic configuration operating on semantically grounded trajectories. Results show that raw event-level learning remains informative, but learned temporal modeling over semantic capability and contextual direct-impact trajectories achieves the highest aggregate point estimates, with the largest margins on the temporal tasks. Ablation analysis indicates that capability dynamics, contextual impacts, and temporal structure contribute complementary information. Shortcut diagnostics indicate that the most direct cross-run global-identifier shortcut is controlled in the primary neural input view, while residual shallow signals remain. Overall, the findings support a bounded methodological conclusion: in controlled synthetic settings, semantically grounded trajectories provide an effective representation for temporal structural reasoning, supporting further investigation of semantic interfaces between event data, structured representations, and temporal learning.

In recent years, the growing complexity of last-mile pickup operations has increased the need for fast and accurate decision-making on logistics platforms. This challenge is fundamentally driven by two key and tightly coupled decision-making processes: order dispatching and routing. Solving them separately overlooks their interdependence, while fully end-to-end learning can be unstable and costly on large, variable-scale instances due to sparse rewards. To solve this problem, we propose an integrated optimization framework which couples a learned routing oracle with real-time dispatching heuristics. For the routing subproblem, we develop a Dynamic-Residual Graph Attention Network encoder with a Look-Ahead Courier-Personalized decoder. For the dispatching subproblem, we develop a routing-oracle-guided dispatching heuristic with local search, where the oracle provides near-optimal solutions to select candidate couriers while retaining real-time scalability. Extensive experiments on real-world datasets from Cainiao Logistics are used to test the performance of our approach, including an offline evaluation and an online rolling-horizon simulation. The experimental results show that our approach outperforms other benchmarks regarding solution quality and solving time, indicating it can effectively support logistics companies in solving real-time and large-scale last-mile pickup problems.

Inferring physical properties such as mass, stiffness, and elasticity from a single image is essential for simulation and embodied AI, yet most existing approaches rely on multi-view reconstruction or physics-based supervision. We introduce SiPhy, a unified framework for single-image physical property reasoning that aligns 3D-aware visual cues, depth with language-based material knowledge. From one RGB image, SiPhy samples pseudo-voxel points, extracts CLIP features, and grounds them to material candidates proposed by a VLM. A part-based contrastive aggregator enforces region consistency, while a heaviness-aware refinement improves thickness and volume estimation for dense objects. Across ABO-500, MVImgNet-100, and PhysXNet-100, SiPhy achieves state-of-the-art single-image performance, surpassing multi-view reconstruction methods by improving mass MnRE by up to 93% (vs. PUGS), reducing density MAE by 35.5% (vs. NeRF2Physics), and lowering Young's modulus error by 23.5%. We further validate SiPhy on real hand-object interaction datasets, demonstrating its potential as a data annotation engine for physical understanding from single-view imagery.

We introduce time-reversed imaging, a new paradigm that infers what just happened in a scene from fading multimodal traces. Instead of extrapolating or interpolating video frames, our goal is to infer past human-environment interactions from residual physical imprints observable in thermal, ultraviolet, and visible spectra. To study this problem, we present TRACE-HEI, the first proof-of-concept dataset for time-reversed imaging, containing synchronized tri-modal video sequences of actions such as sitting, touching, moving objects, and liquid spills, captured across diverse materials and recorded up to three minutes after contact. To establish the benchmark, we propose a multimodal inference approach that extracts structured textual descriptions of detected traces and uses them to constrain a vision-language-guided diffusion model for reconstructing plausible past frames. Experiments show that inferring recent events from fading traces is challenging but feasible when complementary modalities reduce solution ambiguity. This work defines the first computational and experimental foundation for time-reversed imaging, bridging vision, physics, and generative reasoning, and opening new directions for scene understanding beyond instantaneous observation.

Generative social robots (GSRs) powered by large language models offer new possibilities for personalized tutoring in higher education, but also introduce risks related to misinformation, missing transparency, or reinforcing incorrect student responses. Prior work identified knowledge-based design (KBD) requirements that define the informational prerequisites for GSRs to manifest responsible and effective tutoring behavior in higher education. In this paper, we operationalized selected KBD requirements in the Reachy Mini robot platform through system prompting, retrieval-augmented generation, and stateful prompt orchestration. As a result, we present Teachy Mini, a GSR tutoring system that was developed using KBD. To test the system, we conducted a preliminary evaluation study. Participants (N = 24) completed a robot-guided learning session about research methodologies. They learned either with Teachy Mini or with a control version that did not follow KBD principles. Teachy Mini was perceived as significantly more aligned with responsible tutoring behavior than the control robot. Moreover, a manipulation check illustrated that Teachy Mini used personalization, slide-grounded explanations, Socratic questioning, affective support, and learner-anchored feedback more consistently than the control robot. No significant between-condition differences were found in system acceptance, intrinsic motivation, or learning effectiveness, although exploratory analyses suggested a positive effect of KBD on objective learning gains when accounting for learner preferences. Overall, the study offered an initial implementation and preliminary evaluation of KBD for GSR tutoring, indicating that KBD can shape responsible robot behavior and potentially increase learning effectiveness in robot-supported learning.

Open-weight language models from different families exhibit complementary capabilities, motivating their consolidation into a compact student through on-policy distillation (OPD). However, full-vocabulary OPD typically assumes a shared tokenizer, while existing cross-tokenizer methods may discard teacher probability mass or assign it to student tokens with unrelated content. We introduce Byte-Prefix Marginalization (BPM), which re-expresses the teacher's next-token distribution over the student vocabulary in a shared byte space. Specifically, BPM assigns each teacher token's probability to the longest student token whose byte representation is a prefix of the teacher token's bytes, aggregates mass mapped to the same student token, and places otherwise unmatched mass in an explicit residual category. This produces a vocabulary-complete, byte-aligned, and mass-preserving target for dense OPD. The target exactly recovers the teacher-induced byte-prefix marginal when the relevant prefix does not span multiple teacher tokens (a condition satisfied at more than 99% of training positions) and uses a mass-preserving, chain-factorized lower bound otherwise. Across Qwen3-32B, GLM-Z1-9B-0414, and MiniMax-M2.7 as teachers, BPM consistently outperforms current cross-tokenizer methods on six mathematics and programming benchmarks, improving six-benchmark avg@8 by 3.7-6.6 points over the strongest baselines.

Large language models (LLMs) and AI agents have demonstrated strong potential for data integration in zero-shot and few-shot settings. However, they continue to face significant accuracy and cost challenges in enterprise environments due to a persistent knowledge gap. This paper envisions trustworthy, scalable, and cost-efficient integration through knowledge-grounded LLMs and agents operating within a retrieval-augmented generation (RAG) workflow. Here, trustworthiness refers to evidence-grounded, verifiable reasoning, where integration decisions are transparently supported by retrieved knowledge, robust against hallucination, and consistent across tasks. We trace the evolution from classic RAG to GraphRAG and KG-RAG (knowledge graph-based RAG), highlighting how these paradigms bridge parametric and contextual knowledge. Building on this trajectory, we explore the shift toward Agentic RAG, where autonomous multi-agent systems adaptively plan, retrieve, refine, and reason for complex integration tasks. We examine optimization strategies for cost-efficient integration, addressing computational bottlenecks in large-scale enterprise settings. Finally, we outline open challenges and future directions toward building reliable, explainable, and scalable knowledge-grounded integration systems.

Synthetic data augmentation in speech is common practice for linguistic tasks like ASR, but has seen far less work for paralinguistic ones, especially clinical tasks where labelled data is expensive and some patient groups are underrepresented. Voice cloning is one such augmentation approach, but is typically evaluated on speech intelligibility (WER) or speaker similarity (SS) rather than on downstream performance, and it remains unclear whether these preserve the paralinguistic signal such tasks depend on. We benchmark eight voice cloning models on five paralinguistic tasks across public and clinical datasets, showing most preserve signal with modest degradation. We then clone English clinical speech into Japanese and find that training on cloned data outperforms raw cross-lingual transfer for depression and anxiety detection on real Japanese speech, suggesting voice cloning is a promising direction for augmenting clinical speech data in low-resource languages.

Forecasting multiple time-series with high-dimensional covariates presents a core challenge: unifying common temporal patterns while retaining meaningful series-specific information. We introduce Hopformer (Homogeneity-Pursuit Transformer), a two-stage framework that addresses this challenge. In the first stage, we perform a Sparsity Pattern Aggregation (SPA) scheme extracting a common low-variance trend that incorporates the covariates. This acts as a homogenization layer. In the second stage, a LoRA-fine-tuned Transformer models the remaining complex dependencies in the residual. Our method is theoretically grounded. We prove that SPA achieves a near-optimal bias-variance trade-off via an oracle inequality. We also provide generalization bounds for the second stage under dependent time series data. Hopformer sets a new state of the art, improving MASE by an average of 6.56% across synthetic and real-world forecasting benchmarks.

Compressible multiphase flows involving shocks and material interfaces arise in applications such as bubble collapse and droplet breakup, where strong nonlinear interactions produce complex interface deformation, mixing, and multiscale dynamics. Developing reliable machine learning surrogates for these flows remains challenging due to the simultaneous presence of compressibility, sharp discontinuities, and multiphase effects. In this work, we introduce the first large-scale benchmark specifically designed for shock-driven compressible multiphase flows, comprising 2.4 TB of high-fidelity 2D and 3D datasets \footnote{Dataset repo: https://huggingface.co/FluidVerse. Dataset sample videos, metadata.json, inference rollout plots from autoregressive rollout of the trained baselines are provided in the supplementary\_material.zip } featuring shock-induced bubble collapse and droplet breakup. We evaluate diverse surrogate model families on our benchmarking framework: Neptuna \footnote{Benchmarking repo: https://anonymous.4open.science/r/neptuna-A4E3}, including convolutional, spectral, transformer-based, and pre-trained PDE foundation models. Beyond standard MSE training, we investigate composite losses combining MSE with Sobolev, interface-aware, and structure-aware terms, together with adaptive loss balancing using SoftAdapt and GradNorm. Evaluation includes pointwise, spectral, feature-focused, structural, and physics-informed metrics. Results show that no single model performs best across all datasets and metrics, while composite losses significantly improve interface preservation and spectral fidelity. Among adaptive weighting strategies, SoftAdapt provides the most consistent improvements with almost no overhead compared to MSE-only training.

Modeling shared and subject-specific structure in multisubject spatiotemporal data remains challenging, particularly in neuroimaging, where both spatial and temporal patterns exhibit rich variability across subjects. Existing matrix and tensor decompositions provide interpretable factorizations, but rely on fixed multilinear structures or coupling schemes that may limit their flexibility in capturing complex variability. In this work, we introduce a spatiotemporal variational tensor decomposition (ST-VTD) framework that combines a tensor factorization generative model with structured priors to jointly represent spatial maps and temporal dynamics. Spatial factors are regularized to promote a low-rank structure inspired by the LL1 decomposition, while temporal factors are modeled using a learned Long short-term memory (LSTM)-based prior, enabling flexible and adaptive dynamics. Posterior inference is performed using an amortized variational formulation by unrolling iterations of an optimization algorithm, leading to an interpretable and parameter-efficient architecture. The proposed inference framework employs a warm-start strategy based on group independent component analysis, which we found to improve optimization performance. Experiments on a realistic synthetic functional MRI (fMRI) dataset demonstrate that the proposed approach significantly improves latent factor recovery compared with representative classical and probabilistic decomposition benchmarks.

Deep learning models using traditional softmax classifiers have achieved remarkable success in various classification tasks. However, their performance degrades significantly on imbalanced datasets. Although Balanced Softmax is widely adopted as a state-of-the-art rebalancing method, it possesses inherent limitations, such as yielding disproportionately lower testing accuracy for tail classes. To mitigate these shortcomings, we propose the Class-Balanced Softmax (CBS). Rooted in a theoretical Bayesian framework and a heuristic power-law assumption, the CBS is a simple logit adjustment that is computationally inexpensive and easily integrated into existing pipelines. Furthermore, we characterise a fundamental phenomenon in models trained on imbalanced data, termed the preference issue, wherein models exhibit higher training error and a larger generalisation gap for classes with limited data. To quantify this issue, we introduce a novel metric and demonstrate that CBS effectively mitigates the preference issue. Extensive experiments on large-scale benchmarks show that CBS is highly scalable and outperforms existing methods, including Balanced Softmax.

Low-Rank Adaptation (LoRA) is a widely used parameter-efficient fine-tuning method for large language models, but its performance depends strongly on how a fixed rank budget is distributed across Transformer modules. Existing adaptive-rank methods usually rely on local gradient statistics collected during training, which introduces extra memory and computation and overlooks task-conditioned global information flow. We propose IFCLoRA, a topology-aware rank allocation method applied before fine-tuning. Using a small calibration set and a frozen pretrained model, IFCLoRA builds a sparse task-conditioned interaction graph whose nodes represent LoRA-compatible modules. It combines a global information-flow topology prior with local gradient sensitivity to compute Information-Flow Centrality scores, which estimate each module's adaptation importance under multi-hop propagation. Ranks are then assigned once under a global budget. Across multiple models, tasks, and low-rank settings, IFCLoRA consistently outperforms LoRA, AdaLoRA, and EVA under matched training configurations and total rank budgets, while retaining training costs comparable to standard LoRA. On mathematical reasoning with LLaMA 3 8B, IFCLoRA improves over LoRA by 1.36 percent at rank 4 and 1.82 percent at rank 8. Further analysis shows task-dependent, non-uniform rank profiles, indicating that global information-flow structure provides an informative and interpretable prior for low-budget parameter-efficient fine-tuning.

Bayesian optimization (BO) is an optimization method that sequentially proposes the next candidate explainable variables for optimizing target variables by balancing exploration and exploitation. BO is often used under a limited evaluation budget, such as hyperparameter tuning of deep learning. Despite its effectiveness, conventional BO may have poor convergence in practical experimental science where each evaluation is often costly and time-consuming. Recently, BO methods have been proposed that accelerate optimization by using pseudo-experimental data that simulate experimental data. However, when only a limited number of experimental data are available, the generated pseudo-experimental data may be of insufficient quality. In this study, we developed PolyBO to improve optimization time by generating high-quality pseudo-experimental data even when the number of trials is limited. PolyBO performs BO efficiently by generating pseudo-experimental data with an adaptively updated versatile parametric model. This low-capacity polynomial regression model is intended to enable efficient BO even with limited experimental data. PolyBO updates the BO surrogate model with a combined dataset consisting of experimental data and pseudo-experimental data and then performs optimization. Using synthetic benchmark functions with diverse landscapes, we found that PolyBO reduced the optimization time by a median of 42\%. For a real-world material composition optimization problem, PolyBO reduced the optimization time by a median of 96\% compared with conventional methods. Overall, PolyBO achieves efficient optimization in settings where each experiment requires a long time.

Despite the growing use of large language models (LLMs) as creativity evaluators, evidence of their alignment with human evaluations remains mixed, raising the question of when and why their judgments converge with or diverge from human judgments. Across three studies and six widely used LLMs, we addressed this gap by identifying the standards underlying LLM creativity evaluation and examining their downstream implications. Study 1 showed that LLMs generally relied on a narrower subset of human creativity evaluation standards. Convergence with human standards was strongest in the novelty dimension, whereas divergence was clearest in the contextual dimension, which captures social, market, and reputational information. Moreover, each LLM exhibited distinct, model-specific standards that varied substantially in breadth. These differences in evaluation standards were reflected in actual creativity judgments. Study 2 (N = 1,103 ideas) showed that LLM evaluations were moderately correlated with human evaluations, and individual LLMs with broader standards better distinguished ideas humans judged as more versus less creative. Study 3 (N = 1,195) showed that LLMs were less sensitive to contextual information: such information significantly altered human creativity ratings but left LLM ratings largely unchanged. Together, our findings help explain the mixed evidence on LLM-human alignment, showing that alignment depends on the evidence a judgment demands and the standards each model applies. LLMs may resemble humans when evaluations emphasize intrinsic qualities such as novelty, yet diverge when judgments require contextual information. Selecting an LLM evaluator is therefore a consequential decision: different models, applying different standards, recognize different ideas as creative.

In this study, we introduce latent PDE mapping, a broadly applicable physics-informed learning technique designed to enable efficient geometric generalization with sparse training data. Latent PDE mapping pulls back geometry-specific PDE residuals and boundary conditions to a predefined latent geometry via the deformation gradient, thereby enabling the automated calculation of geometry-consistent shape gradients that are missing in conventional physics-informed machine learning formulations. We demonstrate the utility of latent PDE mapping in solving the anisotropic Aliev-Panfilov PDE of cardiac electrophysiology using both physics-informed neural networks and physics-informed deep operator networks. The Aliev-Panfilov PDE serves as a challenging exemplar: a nonlinear, time-dependent PDE benchmark with sharp gradients that are expensive to capture using traditional numerical solvers. To represent the limited data regime, we train the networks using just fifteen geometric samples drawn from parameterized distributions in two and three spatial dimensions. While modest improvements appear for geometries parameterized by affine and shear deformations, latent PDE mapping demonstrates significant benefits on select geometric families, achieving a factor ~4-6 reduction in mean relative L2 error. Furthermore, our results show that the computational cost of applying latent PDE mapping was modest during network training, and negligible at inference. Taken together, our study highlights how latent PDE mapping facilitates the creation of generalizable physics-informed machine learning models from limited sets of training geometries.

Visual anomaly detection requires adaptive representations and reliable decision boundaries, particularly when anomalous training samples are scarce and class distributions are highly imbalanced. Classical kernel-based methods yield principled geometric decision regions but typically operate on fixed features, while deep detectors learn task-specific representations but often fail to provide an explicit margin-aware kernel boundary. In this study, we propose DLM-SVDD, a deep large-margin novelty-detection framework that jointly learns convolutional features and an explicit kernel-based decision boundary. By drawing on the large-margin $\ell_p$-Support Vector Data Description ($\ell_p$-SVDD) approach, the proposed method performs explicit margin maximization and nonlinear slack penalization while adapting the representation to the target task. To train the proposed model, we present an optimization scheme that alternates between a Frank--Wolfe--based update of the convex dual boundary and a CNN update step operating on a smooth margin-violation loss induced by the recovered boundary. To improve scalability, we analyze the efficiency--accuracy trade-offs for different kernel approximation strategies, deriving practical propositions for large-scale anomaly detection. Extensive experiments on multiple standard benchmarks show consistent performance improvements over the baseline and strong overall performance compared with state-of-the-art methods while illustrating that the proposed joint representation--boundary learning scheme remains effective under severe imbalanced class distributions.

Remote sensing multimodal large language models (RS-MLLMs) have improved general aerial-image understanding. However, Earth observation applications require fine-grained scenario specialization, constrained by scarce high-quality scenario data and incomplete capability coverage. We formulate this adaptation as a capability-gap-driven post-training problem and propose filling before advancing (FBA). Rather than relying on single-stage supervised fine-tuning (SFT) over target-domain samples, FBA first fills prerequisite capability gaps before advancing toward scenario specialization. We instantiate FBA for coastal harbor understanding, a representative multi-source scenario, by constructing CPRS (Coastal-Port Remote Sensing), a three-layer supervision dataset coupled with three ordered stages: (1) RS semantic anchoring for overhead-view visual-language alignment; (2) domain-bridge convergence for shared RS priors across target and bridging scenarios under different modalities; and (3) evidence-grounded scenario tuning for downstream performance. We construct HarborEval, an eight-track diagnostic benchmark covering perception, spatial understanding, robustness, and generation. Under comparable training budgets, HarborEval increases from 57.95 with Direct-SFT to 70.29 with FBA on LLaVA-v1.5, and from 81.09 to 83.37 on Qwen3-VL. FBA also outperforms Collapsed-SFT and leads on harbor-related VRSBench/RSVQA subsets and OpenEval. Stage-wise and role-replacement analyses validate progressive gap filling and stage-specific roles. Public examples and release updates for CPRS, HarborEval, code, and trained weights are available at https://github.com/Z0ngL1ng/filling-before-advancing.

Asynchronous reinforcement learning (RL) accelerates large language model (LLM) post-training by overlapping rollout generation with policy optimization, but the resulting stale, off-policy data can destabilize optimization and ultimately cause policy collapse. Existing methods typically retain or discard tokens based solely on the magnitude of their importance ratios, applying the same threshold uniformly across token positions. In this work, we reveal that the natural scale of the importance ratio varies systematically with token entropy. Under asynchronous dynamics, this entropy-ratio scaling dictates two distinct phenomena: at low entropy, the inherent train-inference discrepancy is drastically amplified into substantial sampling noise; at high entropy, in-flight weight updates naturally induce pronounced, legitimate exploratory deviations. Consequently, magnitude-only correction inadvertently admits the amplified noise while strictly masking out the essential exploration triggered by in-flight updates. To address this, we propose the Entropy-Scaled Trust Region (ESTR), which scales each token's off-policy deviation by its local entropy, requiring no auxiliary forward passes or explicit version-switch detection. Across long-horizon agentic tasks and mathematical reasoning benchmarks, ESTR consistently outperforms existing asynchronous methods and achieves the best train-inference consistency. Compared with synchronous GRPO, ESTR attains comparable accuracy while improving training speed by $2.6\times$.

Large language model (LLM) evaluation spans diverse tasks and benchmarks, yet evidence remains organized around tasks rather than the capabilities they probe. This fragmentation limits cross-study comparison, obscures capabilities tasks recruit, and makes coverage gaps difficult to identify. We introduce a multi-layer taxonomy of 14 capability domains and 91 subskills across Primitive, Constructed, and Integrative layers. Human cognitive science guides capability definition and organization, not LLM architecture. Layer assignments draw on developmental precedence and hypothesized functional support, while human-origin constructs are adapted to observable model behavior. To demonstrate operational utility, we screened 31,505 papers from ACL, AAAI, ICML, and NeurIPS between 2023 and 2025 and mapped 15,934 LLM-focused papers through multi-model annotation, consensus, and arbitration. Direct research attention concentrated on Language-Semantic Competence (3,551; 22.3%), Reasoning (3,388; 21.3%), Planning and Decision-Making (2,149; 13.5%), and Perception (1,954; 12.3%), whereas six domains appeared in fewer than 2% of papers. Within domains, the most frequent subskill had a median prevalence of 97.9% and appeared in at least 90% of papers in 10 of 14 domains. Language-Semantic Competence and Reasoning formed the highest-volume pair (n = 1,864; 11.7%; lift = 2.47), whereas Theory of Mind and Social Reasoning and Interaction showed the highest lift among pairs with at least 20 co-occurrences (n = 62; lift = 30.84). By shifting the unit of analysis from isolated tasks to structured capabilities, the taxonomy supports research organization, coverage audits, evaluation interpretation, and testable hypotheses for diagnosis, training, and transfer.

Path planning under partial observability remains challenging because an agent must make long-horizon navigation decisions from only locally bounded observations. Nevertheless, historical trajectories contain reusable experience-guided directional preferences. Classical planners, however, typically solve each instance from scratch and lack an explicit mechanism to exploit such transferable decision knowledge, often leading to redundant node expansions and locally myopic search behaviors. Motivated by this limitation, this paper proposes ImiPath, a prior-guided learning framework that distills reusable spatiotemporal decision priors from demonstration trajectories and uses them as experience-informed directional guidance to bias planners toward reliable and promising search directions under partial observability. Specifically, ImiPath first constructs a local spatiotemporal observation representation, which encodes the spatial information of the local environment and the temporal information of historical trajectories. The SpatioTemporal-Attention Policy Network (STAPNet) then transforms this representation into dicision priors. These priors are further incorporated into heterogeneous planners as directional guidance, biasing the search toward locally promising regions. Extensive experiments demonstrate that ImiPath achieves competitive path quality and improves search efficiency by reducing redundant node expansions under local observability. Additional physical experiments on a magnetic microrobot platform further validate the adaptability and practical deployment potential of the proposed framework.

LLM-based database agents show promise, but differing task scopes, testbeds, and metrics hinder comparison. We identify four gaps between evaluation and production operations: live-environment fidelity (multi-turn read-write interaction with a running database); observation-space scale and complexity (causal diagnosis across thousands of time series, business logs, and concurrent activity); solution-space openness (multiple remediations with different operational trade-offs); and scenario complexity and coverage (faults cascading across internal mechanisms and operational domains). We present DBA-Bench, a benchmark addressing these gaps through production fidelity, outcome-first evaluation, and controlled scenario reproducibility. It uses instrumented PostgreSQL environments with active workloads, persistent state, and multi-source observations; defines success by measurable recovery or fault elimination under safety constraints; and restores snapshots with scenario-specific checks before each run. The benchmark contains 106 scenarios across seven task domains, with two public difficulty labels based on reference-path diagnostic depth and environmental complexity. We evaluate nine baseline groups, including six foundation-model systems, two GPT-5.5-backed database agents, and a Human DBA reference. Across 848 automated runs, Diagnosis, Outcome, and Safe Pass rates are 32.7%, 19.6%, and 12.4%; the best automated baseline reaches 17.9% Safe Pass versus 93.4% for the Human DBA reference. Automated Safe Pass falls from 19.6% on Easy scenarios to 7.6% on Hard scenarios, underscoring the difficulty of safe end-to-end remediation.

AI agents encounter learning opportunities in every episode they run, and discard nearly all of them: the underlying models are frozen at deployment, so an agent that resolves a difficult request today starts from zero when it recurs tomorrow. Yet ordinary operation already produces feedback, in the form of outcome verdicts and after-the-fact corrections. We show that this feedback is a sufficient signal for continual learning when the frozen model is paired with an external memory that distils each episode into retrievable natural-language rules. On the banking domain of $τ$-bench, against a static-RAG control retrieving over the complete policy corpus, learning from the one-bit outcome verdict lifts single-trial success to 1.6$\times$ the baseline, and learning from corrections to 2.6$\times$, converting 22 of the 84 tasks the baseline never solves. The result spans the deployment spectrum, measured on Mistral Large, an open-weights model that organisations with data sovereignty requirements can self-host, and replicated on a frontier model, Claude Sonnet 5. The accumulated memory also transfers: each model, reading the store built by the other, rises above its own no-memory baseline. The harness, protocol, and data are released.

Despite recent advances, self-supervised learning (SSL) models and Joint-Embedding Predictive Architectures (JEPAs) remain susceptible to learning spurious biases in the dataset. These techniques rely on regularization, which prevents representation collapse by enforcing a global target distribution such as a multivariate Gaussian or a uniform distribution on the sphere. However, these global constraints are insufficient to prevent bias entanglement, as task-irrelevant features can still segregate the latent space into distinct sub-regions. While recent approaches like Entangling and Disentangling (EnD) and Fair Supervised Contrastive Learning (FSCL) empirically debias the latent space, we show that they act as partial approximations of conditional distribution matching. To enforce this matching explicitly, we propose Unbiased Open World Regularization (UOWReg), an encoder-only framework. We show that this shift from a global to a conditional objective guarantees statistical independence between the learned representations and the targeted attributes, regardless of the chosen target distribution. We empirically validate this framework across both Gaussian and spherical latent spaces, using statistical measures to enforce these target distributions. While conditional matching successfully mitigates bias with both distributions, we demonstrate that enforcing conditional uniformity on the sphere yields a lower linearprobing classification error. Empirically, UOWReg reduces Equalized Odds violations on the CelebA benchmark while maintaining competitive classification accuracy compared to existing encoder-only baselines. Furthermore, we introduce the Synthetic Engraving Task-a novel setting in which a dominant macro-structure masks a fine-grained micro-signature. We show that UOWReg effectively prevents the subpopulation collapse observed in standard SSL, successfully isolating micro-signatures even when heavily entangled with the global structure.

Accurate pixel-level classification of coronary angiograms is critical for cardiovascular disease assessment, yet the field lacks standardized evaluation protocols. In this work we demonstrate a new benchmark for the assessment of deep learning models which densely classify pixels of coronary angiograms to one of SYNTAX classes (or background). The evaluation covers 24 distinct architectures starting with classic convnets to recent state-space-based vision algorithms. We release CARDIAG - a multi-center, multi-label dataset which we carefully split to reliably compute metrics, accounting for diameter error, overlap, centerline quality and calibration. The data contains SYNTAX labels, binary, uncertainty and segmentation masks as well as intermediate frames together with the selected non-sensitive DICOM metadata. From the multitude of algorithms, we nominate ConvNeXt V2 encoder with DeepLab V3 Plus decoder as the best performing, achieving macro $F_1=0.456$, which we then ensemble with Mamba U-Net and Feature Pyramid Network, for an increased $F_1=0.479$. We demonstrate all the architectures to be well calibrated and determine the generalization of the top 5 methods, together with the data efficiency of these architectures. We highlight the importance of both high-resolution and low-resolution features in encoding. We also demonstrate the model correctness in the context of patient demographic, vessel sides and projection angle configurations. Overall the released benchmark allows for future studies to robustly and rigorously assess the proposals, not only for SYNTAX segmentation, but lesion detection and many more.

Multi-stream robot manipulation policies achieve unparalleled sample efficiency and generalization by modeling actions relative to environmental reference frames. However, existing approaches typically assume these frames to be strictly exogenous. This causal assumption collapses in dynamic settings, such as when a single robot arm manipulates a moving object or when two arms coordinate, where each arm effectively becomes part of the dynamic environment of the other. We propose DynaMAC, a lightweight, policy-agnostic framework that resolves this causal limitation while preserving the sample efficiency, computational speed, and flexibility of multi-stream policies, DynaMAC treats the opposite arm as a dynamic task parameter, thereby providing a unified formulation for dynamic manipulation and bimanual coordination without requiring an explicit leader-follower relationship. To rigorously evaluate these capabilities, we introduce DynaBench, a novel benchmark for robot manipulation in dynamic environments. Across both dynamic environments and bimanual manipulation tasks, DynaMAC outperforms leading probabilistic and generative baselines by over 35 percentage points while requiring 20 times fewer samples. Crucially, DynaMAC generalizes zero-shot from static demonstrations to dynamic environments, substantially simplifying data collection and establishing an elegant bridge toward human-robot collaboration.

Given a database S and a natural language question Q, text-to-SQL systems aim to generate an SQL query that correctly answers Q when executed against S. Currently, popular text-to-SQL benchmarks mostly assume unrestricted access to S; in practice, however, user access is often restricted, e.g., through role-based access control (RBAC) policies. This leads to a potential disconnect between benchmarking results and real-world performance: an LLM with high benchmark scores might perform poorly in an access-controlled environment, by frequently violating RBAC, or rejecting a query q that could be answered with only permitted data in S. Motivated by this, we present a comprehensive text-to-SQL benchmarking framework with realistic RBAC constraints, which features an LLM-assisted workflow that augments existing text-to-SQL benchmarks with plausible user roles and access policies. To do so, we formulate the problem of role synthesis as a structured reasoning process over the database schema, in which the LLM first infers the application context from the schema, and then derives role responsibilities and access scopes consistent with this context. This process is audited by human-in-the-loop quality control, in which domain experts perform metric-guided screening on the generated roles. Besides the augmented dataset, the proposed framework also contains evaluation metrics that identify RBAC-specific failure modes, and disentangle SQL utility from access-control compliance. We apply the proposed framework to several widely-used benchmarks, and conduct a systematic empirical study of state-of-the-art text-to-SQL systems. The results show that many solutions (especially open-weight LLMs) with high benchmarking scores under an unrestricted setting suffer sharp performance degradation once access constraints are in place, due to frequent RBAC violations.

Large reasoning models (LRMs) generate long reasoning traces before producing final answers. While these traces may contain useful signals for hallucination detection, harnessing them is non-trivial because long trajectories often include noisy steps that obscure the cues relevant to truthfulness assessment. In this paper, we identify two prevalent forms of reasoning noises, i.e., irrelevant steps and repetitive steps, and show that both substantially degrade hallucination detection performance. Existing confidence-based scores and naive embedding-based filtering fail to reliably separate noisy from informative steps. To address this challenge, we propose REDE, a novel learning framework for denoising reasoning traces for hallucination detection. Specifically, REDE leverages final-answer attention as an automatic supervision signal to shape the step-level representation space, yielding refined embeddings in which noisy steps can be reliably identified and filtered. REDE can be readily plugged into diverse hallucination detectors by operating on the filtered reasoning trajectory after removing noisy steps. Extensive experiments on multiple reasoning benchmarks show that REDE consistently improves detection performance over competitive baselines.

Autoencoder-based anomalous sound detection is attractive for machine condition monitoring because it can be trained using only normal recordings and yields an interpretable anomaly score from reconstruction error. Most prior work uses spectrogram autoencoders, but reconstructing detailed time--frequency patterns is sensitive to noise and transients, and models can reconstruct some anomalous inputs well, weakening normal--anomaly separation. We propose MemNMF, a constrained reconstruction method that operates on the Linear Predictive Coding spectrum, a compact estimate of the spectral envelope. MemNMF initializes a memory module from an NMF dictionary learned on normal LPC spectra and reconstructs each input as an attention-weighted combination of prototypical normal spectral patterns. Experiments on MIMII and DCASE 2020 Task 2 across multiple machine types and operating conditions show that LPC-spectrum inputs improve a standard autoencoder baseline and that MemNMF yields further gains, with especially strong robustness under noisy, non-stationary settings.

We present Nanbeige4.2-3B, a compact general agentic model with 3B non-embedding parameters. It delivers strong performance across code-agent, office-agent, and complex tool-use tasks while maintaining highly competitive reasoning capabilities in mathematics, coding, and science. Nanbeige4.2-3B is pretrained from scratch on 28T tokens with a Looped Transformer that reuses the layer stack to increase capacity without adding parameters. For SFT data and trajectory construction, we expand the diversity of executable environments, task assets, and agentic scaffolds through real-world deployment and large-scale synthesis. Our RL pipeline applies mixed-mode RLHF over Think and Non-Think responses to improve overall model quality and reduce failure cases, length-controlled reasoning RL to balance accuracy and reasoning efficiency, and agentic RL with outcome and process rewards to stabilize long-horizon training. Extensive evaluations show that Nanbeige4.2-3B outperforms larger models, including Qwen3.5-9B and Gemma4-12B, across diverse agentic benchmarks while remaining competitive on reasoning and alignment tasks. Performance with OpenClaw further supports its use as a compact local personal assistant.

Multi-branch architectures and CNN-Transformer fusion have long been regarded as effective ways to improve vehicle re-identification (Re-ID) by combining complementary representations. In this work, we revisit this assumption in the foundation-model era through a comprehensive empirical study. A single DINOv3-pretrained ConvNeXt trained with a tuned recipe achieves 88.19 mAP on VeRi-Wild Small and 77.47 mAP on VeRi-Wild Large using visual cues alone, matching the strongest protocol-verified metadata-dependent multi-branch baseline. Applying training-free re-ranking further improves performance to 92.38 and 83.68 mAP, respectively. Using this strong baseline together with retrieval-level branch diagnostics, we evaluate whether increasing representational diversity still provides measurable gains. Across both benchmarks, concatenating multiple branches built on a shared backbone changes the best single-branch performance by less than one mAP point while increasing the embedding dimension by 4x, and the resulting representation has an effective rank close to the original feature dimension. We further study cross-backbone fusion using an asymmetric frozen-anchor strategy to combine ConvNeXt and Vision Transformer representations. Despite these favorable conditions, Transformer branches consistently remain 13-15 mAP below the ConvNeXt backbone, and paired per-query bootstrap analysis estimates the largest observed fusion gain to be only +0.11 mAP (95% confidence interval). Our results suggest that, under the evaluated setting, improving a single strong foundation-model backbone together with retrieval-stage re-ranking is more effective than increasing architectural complexity through additional branches or heterogeneous backbones. We restrict our conclusions to single-seed training and one family of foundation models and discuss conditions under which these observations may not hold.

The integration of large language models (LLMs) into the nuclear power industry requires outputs grounded in domain-specific knowledge. This study evaluates a 31-billion-parameter open-weight multimodal model (Gemma 4 31B-IT) on its capacity to apply nuclear knowledge by benchmarking eight model-retrieval configurations against the U.S. Nuclear Regulatory Commission (NRC) Reactor Operator licensing examination. We evaluate 14 Generic Fundamentals Examinations (GFE) from the 2015-2021 March sittings (seven pressurized and seven boiling water reactor exams) using the standard 80% human passing criterion. The base model is compared against configurations utilizing supervised fine-tuning (SFT) on Gemini-distilled chain-of-thought (CoT) rationales, retrieval-augmented generation (RAG) with BM25 sparse retrieval over the U.S. Department of Energy Fundamentals Handbook, and retrieval-augmented fine-tuning (RAFT). Within the retrieval pipeline, we compare fixed-size sliding-window chunking against structure-aware chunking. The SFT configuration with fixed-size chunking RAG met the criterion on 8 of the 14 examinations, outperforming all alternatives, whereas no configuration without fine-tuning passed any. Aggregate accuracy reached 79.7%, with a confidence interval spanning the threshold, and 80.2% on PWR items specifically. Furthermore, two regularities emerged: the preferred chunking strategy reverses depending on the model's training state, and RAFT underperforms compared to standard SFT in matching search environments. These results demonstrate which combination of fine-tuning and search approaches achieves operator-level capabilities.

Counterfactual explanations are a prominent approach in explainable artificial intelligence (xAI), providing actionable guidance on what input changes would alter a model's prediction to a desired outcome. While early methods primarily focused on minimal feature changes, recent work incorporates additional properties such as sparsity, actionability and plausibility. Despite this progress, fair and systematic evaluation remains challenging. Existing studies often rely on different data splits, predictive models, and evaluation metrics, which limits objective comparison across methods. To fill this gap, we introduce CEL (Counterfactual Explanations Library), a unified library and benchmark for counterfactual explanations designed to support consistent implementation and evaluation. CEL includes 18 datasets of varying size and complexity and provides implementations or reimplementations of 14 widely used counterfactual methods. Using this standardized setup, we conduct a comprehensive quantitative comparison across a variety of methods on datasets that differ in size, number, and types of attributes. The evaluation protocol incorporates multiple complementary metrics capturing validity, coverage, sparsity, proximity, and distributional plausibility, including density- and outlier-based measures to assess the realism of generated counterfactuals. To the best of our knowledge, this is the first comprehensive benchmark that systematically evaluates recent counterfactual explanation methods within a unified and reproducible framework. While prior libraries and benchmarking efforts exist in the literature, many are outdated, limited in scope, or lack consistent evaluation protocols. The proposed benchmark aims to improve reproducibility, enable fair comparison, and establish a workbench for the development of future counterfactual explanation methods.

Agent security is widely treated as a question about action content. Defenses ask whether an instruction looks malicious. Benchmarks ask whether an agent performs a harmful sounding action. \textbf{We argue that agent security is fundamentally a contextual problem, and that the current content based framing systematically misdefines it.} A command to ``delete user data'' might be a routine administrative request or a prompt injection attacking production systems, and the content alone cannot distinguish the two. Authorization context can. Across every injection task in AgentDojo and WASP, the same action is one an authenticated user would plausibly request in a routine workflow, which makes the conflation a structural property of evaluating security through content. We operationalize contextual security through four properties that must hold jointly and be evaluated continuously across the agent's trajectory. Source Authorization asks who issued the command. Task Alignment specifies the agent's authorized objective. Action Alignment evaluates whether each action serves that objective. Data Isolation governs information flows across privilege boundaries. Under this reframing, indirect prompt injection becomes a Source Authorization violation. Snapshot benchmarks are structurally incapable of evaluating Data Isolation. Existing defenses are reorganized around the property they actually approximate. The contextual reframing changes which defenses are coherent, which evaluations measure something useful, and which attack patterns evaluation can see at all.

Multimodal Large Language Models (MLLMs) are emerging as core reasoning modules for embodied agents, yet it remains unclear how well general-purpose models can solve long-horizon embodied tasks from a single high-level instruction. We introduce MissionBench, a benchmark for mission-level evaluation of MLLMs in aerial 3D environments. It comprises 120 missions across five simulated 3D environments and four task families. Agents must autonomously plan, navigate, and report outcomes using only egocentric observations and its action history, without aerial-specific fine-tuning. Across 22 open- and closed-source MLLMs, the strongest model succeeds on fewer than 35% of missions compared to 84.4% human performance, highlighting the difficulty of multi-step embodied tasks. Despite large variations between model families, we observe gains from scaling, indicating that larger general-purpose models possess stronger zero-shot embodied capabilities. Our analysis shows that mission-level competence requires coordinating multiple capabilities beyond spatial perception, including multi-step planning and adaptive reasoning. This motivates closed-loop evaluation and highlights both the promise and risk of scaling-driven improvements for embodied AI.

Off-Policy Evaluation and Learning (OPE/L) in contextual bandits is rapidly gaining popularity in real systems because new policies can be evaluated and learned securely using only historical logged data. However, existing methods in OPE/L cannot handle many challenging but prevalent scenarios such as few-shot data, deterministic logging policies, and new actions. In many applications, such as personalized medicine, content recommendations, education, and advertising, we need to evaluate and learn new policies in the presence of these challenges. Existing methods cannot evaluate and optimize effectively in these situations due to the notorious variance issue or limited exploration in the logged data. To enable OPE/L even under these unsolved challenges, we propose a new problem setup of Cross-Domain OPE/L, where we have access not only to the logged data from the target domain in which the new policy will be implemented but also to logged datasets collected from other domains. This novel formulation is widely applicable because we can often use historical data not only from the target hospital, country, device, or user segment but also from other hospitals, countries, devices, or segments. We develop a new estimator and policy gradient method to solve OPE/L by leveraging both target and source datasets, resulting in substantially enhanced OPE/L in the previously unsolved situations in our empirical evaluations.

Energy natural gradient descent (ENGD) aligns parameter updates with the curvature of an underlying function-space energy, but existing formulations assume an unconstrained Euclidean parameter domain. We introduce \EMNGDfull{}, a manifold optimization framework for physics-informed and variational neural PDE solvers whose parameters lie on a Riemannian manifold. EMNGD restricts the energy-induced quadratic model to feasible tangent directions and uses retractions to preserve parameter constraints throughout optimization. Under coercivity, we prove that the push-forward of the undamped EMNGD direction is the best feasible approximation to the function-space Newton vector in the energy metric. We establish coordinate invariance, exact reduction to ENGD in Euclidean space, global first-order convergence with Armijo backtracking, and robustness to inexact tangent solves. For quadratic residual energies and generalized Gauss--Newton pullbacks, the Woodbury identity transfers the tangent system to sample space without changing the direction. Nyström approximation provides scalable sample-space solves with controlled direction error and recovers the exact direction after iterative convergence. On the evaluated neural PDE benchmarks, EMNGD achieves higher accuracy and faster convergence than the compared state-of-the-art baselines. Woodbury preserves the EMNGD direction, while scalable-solver diagnostics quantify the accuracy--cost trade-off of preconditioning and residual subsampling.

Rare-regime discovery in parameterized dynamical systems is an active-search problem: find one verified parameter at which a scientifically defined qualitative threshold is crossed, even when acceptable candidates are rare, nonconvex, or fragmented. We introduce Quantum-Classical Phase-space and Stability-Threshold Search (QC-PHAST), an evidence-gated decision protocol and query-accounting framework for finite candidate libraries. A candidate induces a dynamical object, simulator-derived criticality score, and verified first-hit predicate. Scientific metadata and charged pilot evidence are used to assess whether equation-aware search, scalar-score active search, predicate-only search, or only a query-model comparison is admissible. The quantum row is the inherited Grover/Boyer--Brassard--Hoyer--Tapp (BBHT) unknown-$M$ marked-set query reference; it is not a new quantum-search theorem, materialized circuit, or hardware-speedup claim. The result is a regime map. Direct boundary constructions, geometry controls, online simulator loops, and learned-label accounting further identify when classical structure, false positives, calibration cost, or state preparation erases the query-model margin. QC-PHAST is therefore an auditable protocol for deciding when a finite-pool marked-set reference is informative and when classical or resource-aware search should dominate.

The increasing deployment of large language models (LLMs) has magnified the computational and memory bottlenecks of autoregressive decoding, where low compute intensity and bandwidth-bound kernels dominate inference cost. Weight pruning offers a promising remedy, but existing methods remain confined to either static pruning (SP), which permanently removes redundant weights but lacks adaptivity, or dynamic pruning (DP), which adapts to input sparsity but introduces runtime irregularity. This paper presents SPDP, a unified sparse-inference framework that integrates unstructured SP with input-adaptive DP for efficient LLM inference on GPUs. SPDP co-designs a new Tiled-Column-wise Bitmap Compressed (Tiled-CBC) format and two complementary GPU kernels: (1) a CUDA-core spMspV kernel featuring Hybrid Activation-aware Dynamic Shared-Memory Bitmap Decoding (HAD-SMBD) for fine-grained, runtime activation skipping, and (2) a Tensor-Core SpMM kernel optimized for prefill computation. This joint format-kernel design harmonizes static and dynamic sparsity, maintaining bandwidth-efficient memory access and high compute intensity under both phases of LLM inference. Comprehensive evaluations on inference-optimized GPUs demonstrate that SPDP achieves 1.24x-1.37x average speedup (up to 2.51x) over state-of-the- art sparse frameworks such as SpInfer, while matching. perplexity with up to 25% higher sparsity. SPDP advances the inference efficiency-quality Pareto frontier, showing that unified static-dynamic pruning can deliver substantial throughput and performance-per-watt improvements in large-scale LLM serving

Chain-of-thought prompting improves language-model reasoning by carrying intermediate states across successive computation steps. However, relying on natural language as the only recurrent interface is overly restrictive, since many transient computations do not need to be fully verbalized. Existing latent-reasoning methods remove this constraint by recurrently propagating continuous hidden states. However, these methods pass a dense hidden vector as a whole, without an explicit mechanism for selecting and organizing the information needed by the next reasoning step. This motivates an intermediate interface that remains linguistically grounded without requiring a decoded sentence. We introduce \textbf{J-CoT}, a recurrent reasoning framework built on \emph{J-space}, a vocabulary-indexed coordinate system within the model's hidden representations. Within each cycle, the model computes in its full hidden space. At the cycle boundary, J-CoT expresses the intermediate state as vocabulary-indexed coefficients, carries these coefficients forward as a \emph{J-thought}, and maps them back into the model's hidden representation for the next cycle. J-CoT therefore requires neither a fluent intermediate rationale nor recurrence over the complete hidden state. Under matched backbone and inference settings, J-CoT-Zero matches or exceeds the strongest evaluated latent-reasoning baseline on every benchmark, while J-CoT-Train obtains the highest score across the evaluated mathematical, scientific, coding, and structured path-reasoning tasks.

Test-time scaling through iterative self-evolution with environment feedback, as demonstrated by AlphaEvolve, shows remarkable performance gains. We hypothesize that the success of such evolution frameworks hinges on meta-skills, such as self-reflection with environment feedback, that enable effective multi-round refinement, yet are largely neglected by traditional post-training. To bridge this gap, we present MetaEvolve, a framework designed to develop these meta-skills via a data synthesis pipeline, evolution-aware reinforcement learning (RL), and inference-time evolutionary search. Concretely, we ground MetaEvolve in coding, where program execution provides natural, continuous reward signals beyond binary correctness. Building on these signals, we synthesize evolution trajectories as training data, each containing a current program, its fitness score (combining correctness and efficiency), and a history of prior attempts, and train the model via RL with verifiable rewards derived from test case execution. By training on large-scale code data, we aim to inspire generalizable domain-agnostic meta-skills that can transfer broadly to open-ended problems where such rich training signals are scarce. Across seven coding benchmarks, MetaEvolve outperforms the strongest baseline by 10.01% absolute on in-distribution tasks and 24.12% on out-of-distribution tasks. On open-ended algorithm optimization problems entirely outside the training domain, it further achieves a 46.9% relative improvement. These results demonstrate that explicitly cultivating self-evolution meta-skills offers a principled path toward more capable and autonomously self-evolving AI.

Continual temporal knowledge graph (TKG) reasoning aims to continuously incorporate newly emerging facts while preserving previously acquired knowledge. Replay-based continual learning has achieved promising performance by revisiting historical representations. However, existing methods primarily focus on what to replay, while largely overlooking how replayed representations should be integrated with current ones. Such direct integration often gives rise to two critical forms of representation conflict: \textit{norm domination} and \textit{semantic blurring}, ultimately degrading continual reasoning performance. To address these challenges, we propose MA-DAR (Manifold-Aligned Dynamic Adaptive Routing), a lightweight plug-and-play framework for replay representation fusion. MA-DAR first aligns replayed and current representations onto a shared manifold to alleviate distribution discrepancies. It then employs a dynamic gating mechanism to learn dimension-wise fusion weights, adaptively determining the contribution of replayed and current representations to the fused representation. Furthermore, a polarization regularizer encourages more decisive routing behaviors by discouraging ambiguous gating decisions, resulting in more stable and effective knowledge integration. Extensive experiments on four public continual TKG benchmarks demonstrate that MA-DAR consistently improves the performance of representative TKG encoders while remaining effective under different replay settings. Comprehensive ablation studies and visualization analyses further verify the effectiveness of manifold alignment and dynamic adaptive routing in mitigating representation conflicts and improving continual reasoning.

This technical report presents our approach to Challenge Track~3: SeePhys Pro at the 3rd AI for Math Workshop, where the task is to answer college-level physics questions whose statement and figure may be given partly or entirely as an image. Visual physics problems become substantially harder for large language models when the decisive information resides in a figure rather than in the text, and this modality gap widens as more of the problem migrates into the image. We address the task with a two-stage framework: a visual information extraction stage that re-expresses figure content as solver-readable text to close the modality gap, and a reasoning stage that orchestrates three heterogeneous solvers through multi-agent debate. Our analysis yields two findings: the gain from orchestration comes from reliable answer selection rather than from additional debate, and the value of a figure aid scales with how much of the problem is locked inside the image. The resulting pipeline improves overall accuracy over a single-agent baseline from 0.643 to 0.802 on the public split, and won 1st place on both the public and the private leaderboard (private overall 0.743).

Large language models (LLMs) have become powerful tools for language understanding and logical reasoning. However, they still make mistakes when a problem requires both understanding meaning and following logic. A key reason is that natural-language statements often carry implicit semantic relations before any formal reasoning begins. If these hidden meanings are not properly organized, the model may reach incorrect conclusions even when the subsequent reasoning process appears logically valid. Existing methods improve reasoning through decomposition, symbolic translation, external solvers, or self-verification, but pay comparatively less attention to the semantic structure on which reasoning depends. In this paper, we further investigate how semantic organization influences logical reasoning in LLMs. To this end, we propose HexLogicAgent, a framework that first organizes the meaning of natural-language statements and then guides logical reasoning through structured verification. In our investigation, we also make two observations. First, incomplete semantic representations, rather than deductive inference itself, are a major source of logical reasoning failures in LLMs. Second, explicitly modeling the complete structure of semantic opposition substantially delays the degradation of reasoning performance as logical complexity increases. Experiments on challenging logical reasoning benchmarks demonstrate that HexLogicAgent consistently improves reasoning reliability across multiple LLMs. The core idea is supported by a logical hexagon theory, which explains why a complete structure of opposing meanings is necessary for reliable reasoning.

Developing foundational neural simulators for Partial Differential Equations (PDEs) requires robust generalization across diverse physical parameters and boundary conditions. However, current deep learning approaches largely face a structural trade-off between condition-agnostic deployment and physical fidelity. Purely data-driven operators infer the underlying physics implicitly and thus lack the explicit constraints needed to ensure physically valid solutions across varying domains, rendering the learning problem ill-posed. On the other hand, Physics-Informed Neural Networks (PINNs) enforce rigorous physical constraints but necessitate costly, instance-specific optimization. Furthermore, the massive scale of emerging foundational operators has severely degraded their inference speeds, making them computationally uncompetitive with traditional numerical solvers. To address this bottleneck between condition-agnostic deployment, physical rigor, and inference efficiency, we propose a \textit{Generalized Neural Operator}. By formalizing the classical conditions for well-posedness within neural operators, our framework demonstrates the theoretical benefits of explicitly conditioning on PDE parameters and boundary conditions. To implement this synthesis without compromising computational speed, we introduce three novel architectural components: a parameter-gated mixture of kernels for efficient parameter generalization, a generalized boundary transfer operator that projects arbitrary boundary constraints into a unified latent Dirichlet representation, and a specialized training objective to ensure stability. Extensive experiments demonstrate that our theoretically grounded approach achieves superior generalization across heterogeneous physical regimes while maintaining strict inference efficiency comparable to conventional numerical baselines.

Full self-attention in large language models scales as O(N^2), which limits long-context document analysis to 65,536 tokens and requires costly GPU clusters. The Reduced Interaction Sampling (RIS) inference engine addresses this constraint as a model-agnostic architecture. Without modifying weights, RIS reduces self-attention complexity to O(N log N) using sparse stochastic geometry that fits within commodity memory limits. We validate RIS on Qwen2-1.5B-Instruct across two regimes. In controlled evaluations at 32,768 tokens (where native dense attention serves as the upper bound), RIS-Stochastic at 1% density and 70 ensemble seeds achieves 75.00% accuracy, outperforming the native dense baseline (71.88%), while RIS-Stochastic at 5% density and 10 seeds matches it (71.88%). This demonstrates that sparse attention acts as a regularizer: low density (1%) over multiple seeds filters out sequence-level noise, whereas higher density (5%) reintroduces distractor noise. Under the tightest budget, RIS-Structural reaches 68.75% accuracy at 1% density with just 10 seeds, recovering 75% of the contextual gap relative to the zero-context floor (59.38%). At 65,536 tokens, where dense attention triggers out-of-memory faults, RIS yields retrieval gains of up to 14.06 percentage points over the zero-context floor (51.56%), which is confirmed as marginally significant under McNemar's paired test (p = 0.078 < 0.10). All evaluations run on commodity, unaccelerated CPU servers (16-128 GB of RAM), demonstrating that long-context LLM inference is feasible on standard academic hardware without GPU acceleration.

TRACE-RealWorld addresses a core data-management problem: maintaining an actionable materialized view over a continuously changing physical world when reads of the base state are priced, delayed, heterogeneous, and fallible. Its data-management contributions are a commitment-level validity abstraction for materialized predictions; consequence-conditioned adaptive view maintenance; transaction-style, dependency-scoped compensation for commitments invalidated after authorization; and append-only provenance supporting exact replay. The work builds directly on materialized-view maintenance, adaptive stream synchronization, transaction recovery, sagas, data freshness, and provenance. The end-to-end Flood-SAR evaluation treats sensing as physical data acquisition and measures freshness, verification cost, stale reads, recovery scope, restoration failure, and replayability through six pre-registered questions with held-out seeds. The contribution is therefore not a new predictive model, but a consistency, recovery, and accountability contract for deploying learned world representations as operational data systems.

Coarsening-based training for graph neural networks (GNNs), i.e.\ training on coarsened graphs rather than the original large ones, has become a promising direction for scaling GNNs to massive graphs. However, prior work has been evaluated almost exclusively on \textit{homophilic} graphs, leaving the more challenging \textit{heterophilic} settings underexplored. We show, both empirically and theoretically, that existing coarsening-based training methods suffer significant performance degradation on heterophilic graphs due to inevitable loss of graph information during coarsening. To address this, we propose {\bf A}daptive {\bf C}omplementary {\bf E}nhancement, a plug-and-play, model-agnostic strategy that reintegrates the information discarded in coarsening: ACE learns a projector for re-constructing original node features and applies \textit{anisotropic structural regularization} to embed local heterophily. We further adopt \textit{homoscedastic uncertainty weighting} to adaptively balance the combined training objective of primary coarsened-graph training loss and full-graph auxiliary loss with augmented node features re-constructed by the heterophily-aware projector. Extensive experiments show that ACE drives consistent gains on heterophilic benchmarks while preserving competitive results on homophilic graphs with minimal computational overhead. Code is available at the GitHub repository: https://github.com/vasile-paskardlgm/ACE.

Digital twins have emerged as a foundational technology within the context of Industry 4.0, offering a paradigm for the real-time virtual representation of physical systems. However, managing their growing complexity, particularly in distributed industrial environments, requires intelligent architectures capable of autonomous decision-making, dynamic adaptability, and inter-agent coordination. This systematic review explores the intersection between Multi-Agent Systems and Digital Twins, with a particular focus on predictive maintenance applications in resource-constrained contexts. Through a critical analysis of over 547 papers published in high-impact journals (IEEE Transactions, Nature, Elsevier, MDPI), we establish a taxonomy of existing hybrid architectures, identify persistent technological bottlenecks, and formulate three open research questions concerning: (i) the deployment of artificial intelligence on resource-constrained microcontrollers, (ii) distributed multi-node coordination via lightweight communication protocols, and (iii) the hierarchical orchestration of Digital Twins toward smart factory control integrating residual life estimation and explainable Artificial Intelligence. The results of this analysis reveal that, despite significant progress, no existing system offers an integrated embedded-distributed hierarchical solution that simultaneously meets the requirements of Industry 5.0.

Prior classical-ML learning-curve work fits power laws to tree, linear, and kernel models on tabular data, but at small scale: typically one curve, one team, a handful of cells. We present a distributed classroom-scale replication: 127 students each ran a fixed protocol on 3 assigned datasets, drawn from 18 tabular classification and regression datasets and 6 model families (Boosting, Random Forest, SVM, Linear/Logistic, Ridge, Lasso), yielding 11,536 training runs and 1,648 fitted power-law curves of the form error(N) = a N^(-b) + c. Three findings. (1) Power laws fit: R^2 > 0.8 on 77.7% of cells, with tree ensembles dominating at full data (Boosting 50% of datasets, RandomForest 33%; linear models underperform on classification). (2) Approximate shared exponents within a model family: for 5 of 6 families, a single family-level exponent predicts each family's cross-dataset curves nearly as well as per-dataset exponents (R^2 gap < 0.011), though AIC favors the unconstrained fit and curve collapse is partial (32-58% of points within +/-0.5 dex). We frame this as approximate predictive compressibility, not dataset-independent universality; Lasso fails outright (negative control) and Ridge is fragile under leave-one-dataset-out. (3) Replicator-implementation variance: with random_state=42 fixed, independent re-implementations of the same protocol still differ by mean CV(b) = 0.144 on the fitted exponent -- not seed variance, but the spread induced by unconstrained parts of the protocol (preprocessing, encoding, missing-value handling). We release the aggregated curves, per-cell fits, and a practical data-requirement table for N* to reach target error 0.15.

We study baking documents directly into the weights of a 4-bit Gemma-4-e4b model via LoRA, so a system can answer questions about a corpus closed-book: no retrieval and no context-window budget. Across roughly 100 training runs from single documents to a 99-document corpus, we find that once adapter capacity is adequate, training-data quality is the dominant lever on closed-book accuracy, outweighing LoRA rank, learning rate, and two alternative architectures combined; capacity itself is a hard gate below which no data intervention helps. A single curation pass (shortening gold answers to canonical 1-6 word spans and dropping trivia) moved closed-book accuracy from 57.7% to 85.7% on a 15-document corpus, a larger jump than any architectural change. We confirm a capacity trend (rank must grow with corpus size) entangled with a coupling between rank and learning rate that we initially misdiagnosed. On a 15-document slice we add a real retrieval baseline: the internalized adapter (84.2% recall) beats a BM25-RAG pipeline with a base reader (58.9%) and even a realistic gold-chunk oracle (65.6%) at lower latency. We report the full arc, including three misdiagnoses, as a case study in debugging LLM training empirically.

Constructing causal directed acyclic graphs (DAGs) is a core step in biomedical causal analysis, yet it remains a largely manual process. Analysts must connect study variables to prior literature, evaluate uncertain causal claims, and preserve sufficient provenance for expert review. We present DAGForge, a browser-based system for authoring causal DAGs as auditable, evidence-linked artifacts. Given free-text descriptions of study concepts, DAGForge creates a reproducible literature snapshot, uses an LLM-based reasoning module to generate structured pairwise causal judgments grounded in verbatim evidence excerpts, and assembles those judgments into a constraint-checked graph. Each proposed edge includes confidence estimates, provenance, and a reviewable rationale. The interface supports study specification, progress monitoring, evidence review, graph comparison, adjustment-set computation, and export. In evaluations against both compact benchmark DAGs and reference DAGs derived from published literature, DAGForge achieves high edge recall on the literature-based cohort while retaining verifiable evidence trails absent from LLM-only baselines. DAGForge thus reduces the burden of causal DAG curation while making the resulting assumptions auditable, supporting the design, analysis, and interpretation of biomedical studies.

Modern reasoning models depend on reasoning data, today sourced from human annotations or distilled from stronger LLMs. However, a rich and largely untapped source of supervision lies in expert systems (e.g., game engines, classical planners, theorem provers), which routinely produce near-optimal actions across diverse domains. But these experts are silent: they commit to an action without writing down the chain of thought (CoT) behind it. Recovering that CoT as natural-language reasoning would distill expert knowledge into a student that generalizes beyond the demonstrated actions. We treat it as a latent variable and study how to recover it from the action alone. Our approach, LeAct (Learning to reason from Actions), optimizes this latent variable: the student samples candidate CoTs for each expert action, and we retain those that measurably improve its own probability of recovering the action. Across imperfect-information games at multiple scales and a simulated robotics benchmark, LeAct reaches the solver's numerical floor on small enumerable games. At larger scale, it is $5\times$ closer to the solver than the strongest expert-iteration baseline. At Flop Hold'em ($\sim 10^9$ infosets), LeAct wins head-to-head by $+60$ mbb/g, and on the robotics probe it is the only training recipe that improves on direct imitation. We present a principled framework and the result: expert systems become a categorically new source of reasoning teachers for foundation models.

We investigate whether symbolic regression can discover explicit neural network weight-update rules that outperform standard hand-designed optimizers on small symbolic regression benchmarks. Candidate update rules are represented as fixed-depth symbolic expressions over operands derived from common optimizers, including gradient, momentum, adaptive-gradient, and moment-estimate quantities. Across 30 benchmark/neural network combinations, the symbolic regression procedure found an update rule outperforming the best hyperparameter-tuned established optimizer in 25 cases, with an aggregate MSE reduction of 44.47\% over the improved cases. The discovered rules do not all share a single common symbolic form, but many combine adaptive normalization, momentum-like quantities, nonlinear transformations, and rational expressions. These results suggest that symbolic regression can serve as a lightweight mechanism for discovering compact optimizer variants, while also highlighting the need for larger-scale validation.

LLM agents increasingly rely on external tools, expanding capability while creating a new security boundary: third-party tools may appear benign at the interface level while embedding unsafe behavior in implementation. Existing defenses rely on weak metadata, collapse characterization and policy judgment into a single decision, or use heuristic/LLM enforcement that lacks deterministic, auditable reasoning over task context and multi-tool composition. This paper presents ToolGuardian, a policy-driven framework for securing agent-tool interactions through pre-admission vetting and task-aware runtime authorization. ToolGuardian uses progressive characterization to convert evidence into structured facts: descriptions capture declared intent, system-call traces expose coarse behavior, mock execution reveals observed effects, and source analysis identifies latent behavior. ToolGuardian's core contribution is an Answer Set Programming (ASP)-based declarative policy layer that reasons explicitly over capabilities, effects, task context, and composition. We compare ASP against heuristic and LLM-based policy realizations using identical inputs and output contracts. We evaluate ToolGuardian on 16 MCP-style tools, including 8 malicious variants derived from real open-source tools, and 20 runtime scenarios. For vetting, ASP reaches a deny-class F1 of 0.86 and 88% accuracy using description, syscall, and observed-effect evidence. For runtime authorization, fully specified realizations classify all scenarios correctly, while ablations show that removing compositional and conformance rules substantially degrades performance.

Recent advances in large language models and their rapid adoption across software engineering tasks have made Artificial Intelligence (AI) coding agents an integral component of modern software development workflows. While developers increasingly benefit from these coding agents, their impact on software quality remains insufficiently understood. In particular, how agentic contributions evolve across the software development lifecycle has not been thoroughly investigated. This study aims to characterize agentic pull requests (PR) in comparison to human generated PRs and to examine how their properties change across different stages of the development lifecycle. Using the AIDev dataset, we first analyze how differences in merge rates between agentic and human generated PRs vary over time. We then identify the types of development tasks where AI coding agents are predominantly applied and investigate how these task distributions evolve across development quarters. Finally, we compare a set of key characteristics of agentic and human generated PRs, focusing on their implications for software quality and their temporal dynamics. Overall, our findings provide an empirical and longitudinal perspective on the role of AI coding agents in software development, offering a more nuanced understanding of their benefits and limitations in real-world practices.

Audio deepfake detectors are trained to distinguish genuine speech from synthetic speech and often perform well on standard benchmarks. Yet the same detector that achieves less than 1% error on one dataset can see its error rate increase twentyfold when evaluated on a different dataset. We argue that one contributing factor is speaker-identity reliance: standard training corpora correlate speaker identity with the genuine/synthetic label, allowing detectors to partially rely on speaker-related cues rather than synthesis artifacts alone. We propose the Identity Sensitivity Score (ISS), a per-utterance diagnostic that quantifies how much a detector's output changes across different speaker identity contexts. ISS requires no ground-truth labels at inference time and can be computed from the detector score and a pool of reference speaker examples. Across two detectors and two datasets, incorrectly classified utterances have ISS scores 29 to 52 times higher than correctly classified utterances, and ISS alone predicts misclassification with area-under-curve (AUC) up to 0.954. To test whether ISS actually captures identity-sensitive behavior rather than serving only as a proxy for prediction confidence, we apply voice conversion to 500 utterances and measure the resulting detector-score shift. Utterances flagged as identity-sensitive by ISS respond 19 to 30 times more strongly to this manipulation than utterances flagged as stable. These results position ISS as a practical inference-time diagnostic for speaker-dependent failure analysis in audio deepfake detection.

Previous work has shown that the simple dataflow primitives of the Lustre language allow the natural, semantically unambiguous, and compact representation of machine learning (ML) applications, including models featuring complex conditional execution and recurrent state. The Lustre clock calculus is responsible for the static determination of important properties such as liveness (absence of deadlocks) and static memory bounds. Yet existing clock calculi are tailored for embedded control applications. We show they do not cater for the representation of control patterns commonly found in training algorithms, resulting in cumbersome expressions and inefficient compilation. We propose a conservative extension of Lustre's clock calculus addressing this limitation, thereby facilitating the embedding of ML models in reactive applications.

Roadblocks in Bolivia are a social conflict phenomenon with devastating economic impacts, estimated at losses equivalent to 4% of the national Gross Domestic Product. Despite their recurrence and impact, there is a lack of local predictive systems to anticipate these events for logistical decision-making. This paper presents a hybrid probabilistic forecasting system that integrates time series decomposition (Prophet) with natural language processing (NLP) techniques applied to a six-year corpus of Bolivian news coverage. The methodology employs vector semantic embeddings and zero-shot classification models to capture signals of discursive escalation prior to the materialization of the roadblocks. Using an expanding walk-forward validation scheme applied over 1,762 days and seven forecasting horizons (H+1 to H+7), seven internal configurations and four external benchmarks were compared, including SARIMA and LightGBM. The results demonstrate that the hybrid configuration (Prophet + NLP, C6) consistently outperforms purely statistical models, achieving an AUC-ROC of 0.677 at H+1 and reducing the Brier Score by 10.9% relative to the baseline temporal model (0.220 vs. 0.247), maintaining a statistically significant error reduction across all evaluated horizons ($p < 0.02$). This research validates that the integration of semantic news signals allows for the detection of social tension peaks not captured by historical inertia, providing a technical tool for risk management in critical transport corridors.

Join discovery is a core task in dataset search, enabling users to find columns that can be joined with a given query column. Early approaches focused on equi-joins, but data lakes and open-data repositories often contain columns whose values refer to the same entity but use different syntactic representations. To address this challenge, recent approaches discover semantically joinable columns but face a fundamental trade-off: methods that perform value-level comparisons accurately identify joinable columns but scale poorly to columns with high cardinality; column-level methods that encode an entire column into a single embedding are efficient but do not capture the fine-grained value alignment that determines whether a join is possible. We present MosaicJoin, a value-level semantic join discovery method that balances this trade-off. MosaicJoin achieves scalability through a novel sketching strategy that approximates the joinability of a column pair without having to compare all values. At query time, MosaicJoin scores each candidate sketch using a joinability score at a cost bounded by the sketch size, making retrieval efficient even for high-cardinality columns. A query subsampling operator further reduces online search time with provable accuracy guarantees, enabling robust retrieval for large query columns. Extensive experiments show that MosaicJoin outperforms previously published methods across all benchmarks while running up to 66 times faster than other value-level methods. MosaicJoin requires no training or fine-tuning, and it scales robustly to query columns containing up to 57K values and data lake columns containing up to 1M values.

Document packets, multiple documents concatenated into a single file, are common in government and administrative workflows, yet splitting them into their constituent documents is difficult, especially for low-resource languages. We introduce Khondo (Bangla for split/segment), the first benchmark for document packet splitting on Bangladeshi government forms. Unlike prior English and OCR-text-based datasets, Khondo is bilingual (Bangla--English) and vision-native; where models operate directly on page images. It spans five concatenation schemes, from sequential to fully shuffled, across 14 administrative domains, with ground-truth boundaries, domain types, and page order. Zero-shot evaluation of MLLMs shows they cluster pages into their source documents fairly well but struggle in restoring the original page order once shuffled. To isolate what drives this difficulty, we run two controlled analyses, varying the prompt instruction and then the packet language. Both primarily affect ordering rather than clustering: (a) explicit page-order instructions are necessary but insufficient, and (b) English packets are ordered more reliably than Bangla, making page arrangement the dominant challenge and language a secondary but consistent factor. Khondo establishes page-order reconstruction as a key open problem in vision-based, low-resource document understanding, and provides a controlled benchmark for measuring progress toward solving it. Our dataset and code is available at https://huggingface.co/datasets/Mausul/khondo

Artificial intelligence (AI) has become part of scientific inquiry. Scientists use AI to observe and measure phenomena, to identify patterns in data, and to build models. As AI moves into scientific inquiry, it gains relevance for science education: students should learn how AI is changing scientific practices, ideally by engaging in AI-integrated scientific inquiry themselves. How to design such instruction, grounded in authentic scientific practice rather than taught as a standalone topic, remains an open question. In our vision, which we describe in this article, AI is treated as a set of scientific instruments that students use within the scientific practices described by the Next Generation Science Standards. Each instrument is a genuine scientific tool, pedagogically bounded: its controls are simplified while its core scientific function is preserved. The approach has two aims: engaging students in authentic scientific inquiry, and building an understanding of how AI is used in science and where it can mislead (discipline-based AI literacy, DAIL). In the article, we focus on the investigative core of inquiry, namely observing, analyzing, and modeling, and describe one exemplary AI instrument for each: computer vision for observing, clustering for analyzing, and generative modeling for modeling. We argue that every AI instrument in science education should carry a distinct reflection point that prompts critical evaluation of the AI instrument itself. Finally, we describe how agentic AI, operating across the whole inquiry rather than a single practice, could be represented, arguing that students should first build a foundational understanding of scientific inquiry and AI instruments before relying on agentic AI.

Talking-face (TF) deepfake generation synthesizes photore- alistic facial video from a static source image and an au- dio signal, producing forgeries that current image-based detectors consistently fail to identify. Unlike face-swap ma- nipulation, TF synthesis has no underlying real video from which to inherit physiological characteristics, making re- mote photoplethysmography (rPPG) a uniquely motivated detection modality for this forgery category. We propose a detection framework that extracts per-video rPPG wave- forms via RhythmFormer and trains a suite of lightweight classifiers to distinguish real from synthesized physiologi- cal signals. Evaluated on the TF subset of Celeb-DF++ un- der a strict subject-independent protocol, where test identi- ties are completely separated from training identities, our 1D ResNet achieves an AUC of 0.806 and EER of 27.8%, placing it within 2.4 points of the best published general- purpose detector (Effort, ICML 2025) while operating ex- clusively on the physiological channel. We document a con- trolled reproduction study of DeepFakesON-Phys, the rep- resentative prior rPPG detector, demonstrating degrada- tion from AUC 0.999 on legacy face-swap data to 0.622 on the TF subset of Celeb-DF++. We further show that detec- tion difficulty is strongly method-dependent: AUC ranges from 0.985 (Real3DPortrait) to 0.690 (IP-LAP) across the seven TF generators, with the ranking remaining perfectly stable across all evaluation protocols. This spread reflects an interpretable physiological property of each generator rather than evaluation noise, and constitutes the primary theoretical contribution of the work.

Large language model (LLM) agents routinely cheat on cybersecurity benchmarks, inflating reported pass rates far beyond genuine capability. Prior audits of Cybench found cheating in 0.3-3.4% of traces, implicating only a handful of models. We present a controlled prompt-ablation study across 22 frontier models from 7 providers on 23 Cybench capture-the-flag (CTF) challenges under three prompt conditions (no anti-cheat, standard, severe). All 1,518 task traces were individually audited through a four-stage pipeline combining LLM-as-a-judge classification, programmatic verification, judge-verifier reconciliation, and human review. We find cheating is far more pervasive than previously estimated: under baseline conditions, 37.1% of passes involved cheating, 21 of 22 models cheated, and scores were inflated by up to 5x. Anti-cheat prompts reduce cheat propensity from 33.0% (baseline) to 17.8% (standard) to 8.5% (severe) without degrading, and sometimes improving, solve rates. However, even under the most restrictive prompt condition, eight models still produced cheated passes, four showed backfire effects, and cheating escalated from web search toward infrastructure probing. We introduce the "solve rate" metric (clean passes only) to distinguish genuine capability from cheated outcomes, and argue it should be standard practice in any evaluation where cheating vectors are available. Anti-cheat prompts are an effective and essentially free first layer of defense, but they are not a substitute for environmental controls.

Large language models are increasingly used to simulate human preferences in research and practical applications, raising concerns about validation, misrepresentation, and exclusion. Co-designing agents with the people they represent is a promising way to address these concerns, but participation may also mask the problems it appears to solve. This paper explores that tension through a primarily qualitative study in which 12 participants co-designed personal preference agents in the domain of household energy, via a background survey, co-design interview, and validation survey. Participants engaged readily and mostly came to see their agents as representing them well. Independent validation, however, revealed mixed human-agent alignment, with agent responses markedly more homogeneous, decisive, and abstract than the human sample. I argue that participation and process transparency can act as an "overtrust engine" that promotes trust while concealing systematic misalignment with potential structural consequences at scale. I develop this as a core mechanism in participatory preference agent design, treating individual alignment not as a fixed state but as an enacted process.

Large Language Models (LLMs) are increasingly used in database-backed applications to classify tuples, filter records using semantic predicates, extract structured attributes, and enrich query results. Yet the prompt that start these computations are typically stored outside the DBMS in unstructured formats, making them invisible to query execution, metadata management, and optimization. Drawing on Stonebraker's QUEL as a Data Type and the principles of reflective programming, this paper introduces PromptDB, a database system that treats prompts as tuple-level database values. PromptDB provides a logical PROMPT datatype whose values store a template, bindings to tuple attributes, model metadata, and task metadata. Relations may contain PROMPT attributes directly in base tables, or expose them through views over joined tuples. Users query prompt-valued attributes through generated evaluation views, while the system internally renders, rewrites, optimizes, and executes prompts through an EVAL operator. Making prompts database-visible creates a new optimization space. The key idea is to bring query-optimizer thinking to prompts: just as query optimizers exploit database metadata to rewrite SQL plans, PromptDB exploits database metadata to rewrite prompts. We evaluate PromptDB on synthetic and real-world data workloads across different tasks. The results show how database-guided rewriting improves output validity and yields favorable cost-quality trade-offs compared with static, manually written prompts.

Red-team evaluations of AI models support some claims and not others, and the boundary between the two is calculable rather than merely a matter of judgment. We define the evidential ceiling of an evaluation as the largest factor by which one result can move belief under a fixed testing budget, derive it in closed form for the benchmark null result, and use it to locate that boundary exactly. We find that above a calculable harm rate, a benchmark of modest size certifies a category to a stated evidentiary standard, and a clean sheet is then the stronger of the two possible observations, outweighing a single reproduced failure. Below that rate, no passive benchmark of feasible size provides the specified evidence of safety under the fixed scoring rule and approximately independent trial structure. The crossing between the two regimes has a closed form. The bound is not specific to benchmarks: written in terms of a procedure's hypothesis conditioned elicitation rates, it covers adaptive and automated red teaming as well, and shows that discrimination between the hypotheses rather than attack success is what determines evidential worth. Auditing eight evaluation suites against the boundary, we find that current benchmarks are adequate for high-frequency harm categories and several orders of magnitude short for rare, catastrophic ones. Safety benchmarks are not uninformative. They are informative about a specific and computable set of propositions, and the discipline they need is to state which.