Papers

We test the performance of agents for automated network intrusion response in a cyber range intended for human operator training. The range implements an emulated networking environment with a variable network topology, red-team emulation and simulated user agents. The goal of the defensive agents is to prevent hosts in the network from being accessed by the red-team agent, while minimizing the availability costs induced from defensive measures. Alerts are generated using a SIEM platform and mapped to a data modeling language used by the agents. We test a combination of heuristic agents and policies learned using reinforcement learning. The learned policies are optimized to minimize the combined cost using a cyber attack simulator modeling the network. We found that the reinforcement learning agents were overall more efficient at defending the system than the heuristic policy, and that the performance depends highly on the policy of the adversary in combination with the simulated users.

State Space Models (SSMs) have emerged as a compelling alternative to Transformers, enabling sequence modeling with constant memory and linear compute. Although SSMs exhibit reasonable performance and favorable computational characteristics, they continue to lag behind Transformers on tasks that require in-context learning and precise retrieval, slowing their adoption for large-scale language modeling. In this work, we demonstrate that both the success and failure of SSMs in these domains can be explained by studying the role of the gating mechanism, a prevalent component in modern recurrent networks. Specifically, we show through theory and experiments that this gating mechanism causes SSMs to first learn an in-weights "memorization" solution, while delaying, or even preventing, convergence to a correct in-context learning solution. Importantly, this happens even in cases where there are no fundamental limitations due to the architecture or its memory capacity. On the other hand, we find that gating is often beneficial for improving generalization to long sequence lengths. Our results illuminate the crucial role of the gating mechanism in shaping both the training dynamics and generalization of SSMs, and provide a basis for understanding and improving linear-time models.

Transformers and state-space models (SSMs) are the two dominant families of sequence models, and a central open question is how far the analytical knowledge built for transformers transfers to SSMs. We address this through the lens of layer importance which underpins compression, selective fine-tuning, and interpretability across both families. We decompose layer importance into two distinct notions. Necessity captures how much the pretrained model depends on a layer's existing contribution, measured by the loss increase from bypassing it. Plasticity captures where the model absorbs new information during fine-tuning, measured by the magnitude of task-specific weight updates. Our analysis reveals that the two families behave fundamentally differently: in every evaluated residual transformer up to $14$B parameters, Necessity and Plasticity anti-align across depth, whereas in the evaluated Mamba-style SSMs they point to overlapping regions. The sign of this alignment also predicts downstream adaptation behavior. In the evaluated transformers, concentrating updates in the most plastic layers increases catastrophic forgetting, while this tier-dependent effect disappears in the evaluated Mamba-style SSMs.

Exploring the chemical space of flexible molecules remains challenging because the vast number of possible compounds and conformations, together with the increasing cost and limited generalization of 3D generative models for larger and more complex molecules, restrict access to unexplored chemistry. Here, we introduce QALPA ("Quantum-Aware Learning for Property-space Augmentation"), a property-guided generative framework that combines an E(3)-equivariant diffusion model with active learning and efficient quantum-mechanical (QM) methods to iteratively explore targeted QM property manifolds. By coupling generation with physics-based evaluation, QALPA improves molecular sampling and model reliability in sparsely populated regions of chemical space. Our results show that training on complementary QM datasets spanning both small (QM7-X) and large (Aquamarine) drug-like compounds enables accurate molecular generation across a broad size range, improving transferability beyond the training distribution for complex property manifolds involving both extensive and intensive properties. As a proof of concept, QALPA coupled with the machine learning-augmented tight-binding method EquiDTB efficiently augments alloQM, a QM dataset introduced in this work, comprising 6,253 conformers of allosteric drug molecules, by populating sparse regions of the property landscape defined by the many-body dispersion energy and HOMO-LUMO energy gap. These results demonstrate that the integration of generative AI with efficient ML/QM methods offers a practical pathway toward augmenting sparse QM datasets and sustainably expanding the exploration of chemical space for molecular discovery.

Scientific research increasingly relies on large, heterogeneous data sources, motivating interest in retrieval-augmented generation (RAG) systems that provide natural language access to scientific knowledge and research workflows. Researchers are exploring the viability of these systems as natural language interfaces for document search and for generating analysis code and pipeline components. At the same time, concerns about data privacy and control over research infrastructure have motivated interest in open-weight models and open-source deployments hosted within research institutions. In astronomy, this development follows a long history of computational infrastructure development, from archival databases and Structured Query Language (SQL)-based systems to large language model (LLM)-assisted research tools. This paper presents a domain-expert evaluation of faithfulness for AquiLLM, an open-weight, offline RAG-LLM platform designed to support scientific research groups in the use and preservation of tacit and formal knowledge. We define faithfulness as the extent to which generated responses remain grounded in retrieved scientific context without unsupported claims or omissions. We report results from an astronomy case study evaluating AquiLLM across retrieval and scientific analysis tasks. AquiLLM performs most reliably on explicit retrieval-oriented questions grounded in the RAG collection, while faithfulness degrades for queries requiring synthesis or ambiguity resolution. These results highlight both the promise and limitations of open-weight RAG-LLM systems for scientific research and demonstrate the importance of domain-expert evaluation beyond standard benchmark leaderboards.

Resume screeners must infer job-relevant competence from resumes whose presentation can vary substantially in wording, structure, stylistic polish, and document extraction quality. Ideally, such surface variation should not change decisions when the underlying qualification evidence is unchanged. We introduce a controlled audit of this property, constructing occupation-grounded candidate profiles at controlled competence levels and rendering each profile into multiple resume presentations. A deterministic validation gate excludes variants that alter the underlying evidence before scoring. Across six open instruction-tuned LLM conditions, we find a clear disconnect between screening validity and presentation stability. Llama-3.1-8B with its native chat template achieves the strongest validity ($0.781$) yet reverses $29.6\%$ of matched pairwise decisions under competence-preserving presentation changes; Mistral-7B-v0.3 reaches validity $0.644$ with a $41.4\%$ flip rate. Native chat formatting improves validity for several chat-tuned models but does not remove this instability. These results show that resume-screening evaluations should assess not only whether a system identifies stronger candidates, but also whether those decisions remain stable when the same competence evidence is presented differently.

De-identified résumé screening assumes that redacting explicit fields prevents ethnocultural inference; however, recent audits attribute residual leakage to declared languages. We investigate whether eliminating language fields resolves this leakage across nine open-weight models and 620 counterfactual résumés. By holding language attributes strictly identical, we isolate unstructured prose across five ethnocultural conditions and three cue-salience tiers. Target-group recovery averages 0.757 overall and saturates at 1.000 under high salience, demonstrating that non-language prose sustains demographic inference. Crucially, models diverge only under faint cues (0.086-0.690), establishing salience as an essential evaluation axis. Furthermore, pairwise LLM-as-a-judge outcomes are highly sensitive to evaluation design: forbidding ties yields an apparent selection-rate ratio of 0.39 alongside strong position and content effects, whereas permitting ties produces near-universal ties for most models ($\ge94\%$). Downstream scoring shows only very small between-condition differences, highlighting the need to distinguish demographic signals recoverable from résumé content from effects introduced by the evaluation protocol.

Research data repositories are essential infrastructure for scientific inquiry and for ensuring that datasets follow FAIR (Findable, Accessible, Interoperable, and Reusable) principles. However, repository reuse depends on the quality and completeness of geospatial and thematic metadata, which researchers generally provide voluntarily. Given limited curation resources, it is unsurprising that even Harvard Dataverse, the world's largest general-purpose research repository, contains many incomplete metadata records. Missing fields represent lost information and reduce interoperability. We find that datasets with more missing metadata receive fewer downstream citations and have fewer resolvable connections to other datasets. The implications are particularly important for geospatial datasets: only 0.3% of research datasets include a bounding box, and most represent archival points rather than complete geographic shapes. Our analysis shows that geospatial metadata helps connect concepts across disciplines. After embedding Harvard Dataverse datasets in a metadata knowledge graph, we find that datasets are twice as likely to connect across scientific disciplines through shared geospatial metadata as through keywords. This suggests that geographic metadata is a more reliable basis for cross-disciplinary interoperability than keyword vocabularies, which often remain discipline-specific. We train and fine-tune a small language model using datasets from Harvard Dataverse. Through geospatial metadata enrichment, we increase the share of datasets from different disciplines connected through metadata elements from 58.5% to 63.2%.

AI policies introduced by software organisations to mitigate LLM-related risks such as sensitive information leaks and unauthorised usage are not useful if software developers do not engage with them. We draw on 19 software developer interviews to show how AI policies help and hinder developers. We suggest approaches to support managers and decision makers with a developer-centric approach to introducing AI policies.

Traditional catastrophe (CAT) risk models rely on costly manual construction to generate extreme weather scenarios, an approach largely unchanged since the 1990s. As climate extremes intensify, this creates mounting challenges to the entire risk transfer chain. This study proposes the TAISE framework, which repurposes AI weather forecasting models to produce coherent extreme weather sequences at a fraction of traditional costs. Through self-iterative generation, the framework produces continuous global atmospheric fields from which extreme events emerge. A proof-of-concept experiment demonstrates an order-of-magnitude reduction in computational cost compared with conventional methods, while capturing temporal continuity and cross-regional correlations absent in snapshot-based approaches. These findings suggest a pathway toward democratising catastrophe risk quantification and enabling dynamic, comprehensive portfolio assessment for insurers, reinsurers, ILS fund managers and public-sector risk managers.

Electrocardiogram (ECG) delineation identifies the boundaries of P waves, QRS complexes, and T waves, providing structural annotations that can guide AI models in learning to interpret ECGs. However, training accurate delineation models requires manual annotations that are scarce and time-consuming to obtain. Recent work addresses this limitation through semi-supervised learning (SSL), but the design of the architecture, particularly the decoder, has received less attention. To this end, we propose R-U-Net, an ECG delineation model that pairs a ResNet-18 encoder with a U-Net decoder. On SemiSegECG, R-U-Net outperforms the strongest evaluated ResNet-18 + fully convolutional network (FCN) head baseline in each of the 16 in-domain settings by 3.3-13.0 mIoU and achieves 82.6 mIoU in the cross-domain setting, an improvement of 8.1 mIoU. Controlled ablations show that decoder design contributes more to performance gains than the evaluated SSL methods, motivating further exploration of architectures for ECG delineation. All code is open-source at github.com/ELM-Research/ECG-Delineation.

We present a framework for evaluating data-science agents on live, continuously updated data using executable ground truth and format-agnostic factoid scoring. Consider this example query: "what were last week's audience sizes"—the reference answer changes as the underlying data changes, so static references become outdated and standard LLM-as-a-judge pipelines cannot verify responses against a fixed ground truth. Our central contribution, ground-truth-as-code, encodes each expected answer as an executable reference function that recomputes the answer directly from live data at evaluation time, ensuring the reference remains consistent with the system it describes. We combine this with a factoid-level, format-agnostic judge that decomposes both the agent's response and the computed ground truth into atomic claims and scores precision, recall, and accuracy over them, irrespective of the response format (prose, list, table, HTML, etc.). The approach is applicable to agents whose expected outputs can be expressed as executable data computations. We validate the framework through a human–LLM agreement study on an internally developed machine learning skill deployed in production, using a synthetic database constructed to reproduce production schemas and entity relationships. Relative to a natural-language ground-truth baseline, our method achieves a 29% improvement in the Matthews Correlation Coefficient (MCC)—a class-balanced measure of agreement between expert annotators and LLM-as-a-judge predictions—and a 16% reduction in token consumption per test case, while a self-directed baseline lacking explicit ground truth is anti-correlated with human judgment. Agents that perform multi-source data integration and computation over non-stationary data are routinely deployed in industry; we propose ground-truth-as-code as a practical methodology for their evaluation.

Agentic large language model (LLM) systems rely on long interaction histories to preserve instructions, tool states, intermediate decisions, and unresolved dependencies, but unrestricted context growth increases computational cost and can reduce efficiency. This study evaluates protocol-preserving context trimming as a reliability-constrained approach for multi-step agentic workflows. Five trimming strategies - recency-based, relevance-based, summarization, protocol-aware trimming, and adaptive budget guardrails - were compared across retained-context levels and workflow-complexity classes using task success, protocol adherence, valid tool calls, token savings, latency reduction, cascading failures, and critical context thresholds. Conventional strategies achieved about 60% mean token savings but lower task success (66.6-77.3%) and protocol adherence (85.5-88.6%). Protocol-aware trimming improved task success to 92.2%, while adaptive guardrails achieved 96.0% task success, 96.3% protocol adherence, and 1.0% cascading failure with 56.0% mean token savings. Retained-context budgets of 25% or less increased failure odds 10.92-fold relative to budgets of 50% or more (p < 0.001). Protocol-aware trimming produced 5.24-fold greater odds of successful completion than conventional methods under aggressive budgets, while adaptive guardrails further increased success odds 2.11-fold versus fixed protocol-aware trimming (p < 0.001). Critical context thresholds also increased with workflow complexity. These findings indicate that reliable context reduction depends more on preserving protocol-critical state than on maximizing token removal, and that adaptive guardrails can improve efficiency, scalability, and reliability in long-horizon agentic systems.

Privileged on-policy distillation improves multimodal reasoning by allowing a teacher to evaluate student trajectories using rich, training-only visual evidence. Both models score these trajectories while conditioning on the same student-generated prefix. When a student misinterprets an image early in a response, this accumulating erroneous rationale eventually pulls the teacher away from its visual evidence. The teacher and student converge on the same hallucination, causing standard cross-model supervision to collapse precisely where correction is most needed. We find that the teacher's visual corrective preference is not lost under this misleading agreement. Comparing the predictions of the identical teacher given the real image and a visual null reveals that the privileged evidence still pushes the model toward the correct interpretation. We introduce OPD-Aha, which reconstructs the distillation target directly from this isolated visual preference rather than relying on the fragile teacher-student discrepancy. This reconstructed target aggressively suppresses continuations that contradict the image. Trained with this objective, students learn to naturally interrupt their own flawed reasoning with reflection tokens such as wait and actually. After reflection, subsequent generation relies less on the accumulated erroneous text and more on the visual evidence. Correcting these trajectories mid-generation fundamentally alters the reasoning process, yielding broad and consistent improvements across diverse fine-grained perception and complex multimodal reasoning benchmarks. Our code and models are available at https://github.com/Echochef/OPD-Aha.

Recent work by Doshi and Hauser (2024), Bisbee et al. (2024), and Xie et al. (2026) raises concerns that outputs from large language models (LLMs) tend to be under-diverse: they repeat or resemble one another more often than responses from the population they are meant to represent, a phenomenon known as mode collapse. In this work, we show that whether mode-collapse, or its opposite, occurs depends on the specific model and dataset used. Further, with sufficient supervised fine-tuning (SFT) data, LLM output diversity converges toward that of the target distribution from which fine-tuning data are sampled. To quantify this comparison, we measure the probability that two responses sampled independently from the same fixed prompt coincide (collide), or their expected similarity under a kernel. We derive a bias-variance decomposition of the expected gap between the model's and target's collision probabilities, showing that SFT is not inherently biased toward mode collapse or its opposite: finite-sample SFT can leave a model either under- or over-dispersed, depending on the model and dataset. Finally, we show that the absolute gap is bounded by the square root of the Kullback-Leibler (KL) divergence from the target distribution to the model. Consequently, a model sufficiently close to optimal under population cross-entropy cannot exhibit arbitrarily miscalibrated diversity. We test the decomposition and the bound in three experiments: small transformers on synthetic languages, four LLMs fine-tuned on human surveys, and these LLMs fine-tuned on CodeNet, a dataset of human code solutions. More target data moves model diversity toward the human (or synthetic target) level in all experiments, consistent with our theoretical predictions. These results show that diversity miscalibration can arise from finite-sample error and shrink as SFT better approximates the target distribution.

Diffusion large language models (dLLMs) offer a promising parallel decoding paradigm as an alternative to autoregressive generation through iterative unmasking. However, dLLMs typically require many steps before token confidence reaches the decoding threshold, resulting in inefficient inference even with block-wise KV caching. To accelerate dLLM inference, we for the first time propose an "early-bird (EB)" decoding framework, motivated by the observation that tokens with similarly low entropy tend to cluster and can be jointly decoded earlier, before reaching the confidence threshold. In particular, our EB-Decode framework integrates two key enablers: (1) a learnable network that adaptively groups tokens with similar uncertainty into variable-length blocks, rather than relying on fixed block sizes; (2) a position-aware sampler that learns to unmask tokens in parallel using fewer decoding steps within predicted variable-length blocks. Both components are developed without modifying pretrained dLLM weights and can therefore be directly deployed as plug-ins during serving, with negligible training and inference overhead. Extensive experiments across three models and four benchmarks consistently validate our observation and the effectiveness of EB-Decode, achieving 3.53-18.76$\times$ higher throughput than the vanilla decoding method and up to 1.58$\times$ higher throughput over the strongest baseline, Fast-dLLM, with comparable accuracy.

Satisfiability Modulo Theories (SMT) solvers are foundational to software verification, program analysis, and compiler testing, particularly over the theory of Quantifier-Free Floating-Point (QF_FP). While recent optimization-based SMT solvers have successfully applied gradient descent to continuous relaxations of logical formulas, they are fundamentally bottlenecked by gradient domination, a phenomenon where a small subset of difficult clauses hijacks the optimization trajectory, preventing the solver from satisfying the broader formula and trapping it in local minima. To overcome this, we present GradSAT, a novel framework that bridges optimization-based SMT solving with Multi-Task Learning (MTL). GradSAT reformulates the constraint satisfaction process by treating each SMT clause as an independent MTL task. By applying dynamic gradient normalization (GradNorm), GradSAT actively balances the gradient magnitudes across all clauses at runtime, systematically penalizing dominant gradients and accelerating lagging clauses to ensure uniform convergence. GradSAT implements this through a highly optimized, two-stage hybrid pipeline. First, a GPU-accelerated PyTorch backend leveraging symbolic compilation and operator fusion navigates the continuous relaxation to a high-quality basin. Second, the candidate assignment is handed off to a bit-precise local search engine to rapidly resolve the exact, rigorous assignment. By stabilizing the continuous search dynamics, GradSAT mitigates the brittleness of prior gradient-based solvers and provides a robust, highly parallelizable architecture for complex constraint solving.

Complex problem solving depends on acting effectively and understanding how a system works. AI advice may support these outcomes unequally. Two preregistered experiments compared participants managing a simulated clothing factory with and without an LLM advisor. Across studies, AI-supported participants reported greater confidence and understanding with less effort. In the first study (N=200), assistance increased company value but produced no detectable prediction-accuracy difference. After withdrawal, previously supported participants outperformed controls when decisions were scored against repeating previous choices, but not default settings. Within the AI-supported group, more frequent recommendation alterations predicted better unaided performance. In the second study (N=198), AI-supported participants went bankrupt less often and showed a small knowledge advantage in the registered analysis, largely associated with remaining solvent. More frequent recommendation alterations predicted higher knowledge within the AI-supported group. Applied HAI evaluation should assess users' understanding and independent capability alongside the performance achieved with AI support.

Video diffusion transformers depend on temporal attention to coordinate information across frames, yet nearly everything known about this mechanism comes from analyzing trained models, so when and where temporal-attention structure forms during training remains poorly characterized. Population averages can also hide it, since a few specializing heads and a diffusing majority cancel in the mean. We therefore conduct a checkpoint-resolved census of every temporal-attention head across nine Open-Sora STDiT training runs spanning three model scales (306M to 1.03B parameters), scoring each head with an entropy-normalized measure of cross-frame attention concentration (CFAC) under a preregistered change-point and effect-size selection rule. The census reveals the sparse picture that averages obscure. Aggregate CFAC is flat or decreasing in every run, while a small minority of heads, roughly 4–13% in full-grid runs, develops pronounced concentration. Across seeds, the reproducible signal is positional but block-level. Selected heads repeatedly arise in the first temporal block, whereas individual head coordinates do not reproduce once block membership is accounted for. Among the analyzed 760M selected heads, attention maps converge to a small repertoire of local frame-routing motifs, self-frame diagonals and adjacent-frame bands, even when the responsible coordinates differ across runs. Correlation and ablation analyses do not establish a causal link to generated video quality, and we bound our claims accordingly. Beyond this STDiT family, the study contributes a transferable methodology. Checkpoint-resolved, per-head analysis under fixed selection rules can expose sparse temporal organization in other factorized video diffusion transformers and, with adapted routing metrics, in joint spatio-temporal architectures.

We present Open-Qwen-Music, an open reconstruction of Qwen-Music and a fully specified research system for text-to-music generation that couples LLM-based semantic composition with diffusion-based acoustic rendering. The system comprises a 25 Hz single-codebook music tokenizer, a 3B-parameter autoregressive Music LLM, and a diffusion renderer producing 48 kHz stereo audio, following the cross-module interfaces reported by Qwen-Music. The strongest systems of this design remain closed, and prominent open music-generation projects release weights and inference code without their training corpora or end-to-end training implementations. This limits independent and controlled study of how information loss and prediction errors propagate from semantic representation through autoregressive planning to acoustic rendering. To our knowledge, Open-Qwen-Music is the first fully open release of an LLM-composition-plus-diffusion-rendering text-to-music system. Beyond model weights and inference code, the release includes the training datasets and provenance manifests, complete data-processing, annotation, training, inference, and evaluation pipelines, configurations, and pretrained weights for every learned module. Artifact manifests bind the identities of these artifacts across the complete workflow. Together, these artifacts establish a reproducible implementation of the modular architecture and provide an empirical basis for component-level analysis and future evaluation. We present the system as a transparent, executable research baseline and a starting point for the community, not as evidence of quality parity with Qwen-Music. Open-Qwen-Music is an ongoing effort, and we will continue to improve its generation quality, controllability, and robustness. All release artifacts are available at https://github.com/biang15343100-source/Open-Qwen-Music.

Historical manuscripts remain largely absent from modern vision-language benchmarks, leaving open how well multimodal large language models (MLLMs) handle culturally diverse, degraded, and non-Latin document images. We introduce PalmLeaf-VQA, a multi-script visual question answering benchmark for historical palm-leaf manuscript understanding across South and Southeast Asian traditions. PalmLeaf-VQA contains 923 curated manuscript images and 7,384 question–answer pairs from eight collection groups: Balinese, Grantha, Jathakam, Kambaramayanam, Kannada, Khmer, Sundanese, and Tamil. Unlike recognition-oriented resources, the benchmark targets manuscript-aware visual reasoning over preservation-relevant cues, including physical condition, line structure, material and coating, binding holes, margins, symbols, drawings, and localized visual artifacts. We evaluate recent proprietary and open-weight MLLMs under open-answer and constrained-answer prompting and provide fine-grained analysis across collections, question categories, and task types. The strongest evaluated model reaches only 58.00% exact-match accuracy on the held-out test split, revealing substantial limitations in current MLLMs for rare-script, degraded-layout, and preservation-oriented document understanding. PalmLeaf-VQA provides a standardized benchmark for advancing culturally grounded and layout-aware multimodal document analysis.

The Isabelle/HOL dataset of Benzmüller and Scott's study of Gödel's ontological argument and Scott's variant (Monatshefte für Mathematik, 2025) is carried to Lean 4 and from there back to the automated provers, as a benchmark independent of either proof assistant. The port covers all thirty theories, structure and names preserved: 548 statements compare identical as parsed, every named result is proved again, and five results the original reports without replaying them are proved here. For every theorem, #print axioms gives the postulates its proof consumes: Scott's necessary existence and modal collapse need only a symmetric frame, confirming that KB suffices. The benchmark, in TPTP THF and SMT-LIB, turns the steps of an argument debated in philosophy into 294 theorems, alongside 45 statements the original refutes or leaves open, ten left open there. Five THF provers, and cvc5 on SMT-LIB, prove 227 theorems within ten seconds on one core and 232 within sixty, and none proves any of the 45. E and Leo-II solve the most, although Leo-II's calculus has been unchanged for about a decade and was only repaired and modernised here, as release 2.2. Vampire, whose later version won the higher-order division of CASC-30, solves the most in no configuration. Only E and Leo-II are measured in their own automatic mode: Zipperposition proves 101 in a single mode and 213 with its developers' portfolio, Vampire 174 without options and 209 with a higher-order schedule that its CASC mode does not select, and Leo-III 159 alone and 177 with E as partner.

Medical vision-language models can propose how urgently a skin lesion should be reviewed, but the local service retains authority to accept or replace that proposal under referral policy, capacity, and locally held patient context. Proposal quality and selected-action quality are therefore distinct evaluation targets, and benchmark evidence transfers between them only when local review preserves the expected action score. We introduce AuthEval, a logging and evaluation framework that records both actions, scores the selected action under declared local criteria, and, where feasible, scores the declined proposal under the same rule. It reports the resulting authority gap only when the record supports it. Because the gap is the product of the proposal-change rate and the mean score change on changed cases, that rate alone determines neither its magnitude nor its sign. On ISIC 2019, with MedGemma and simulated local review, two constraint regimes with similar change rates produced an optimistic image-equal gap under capacity ($+0.744$ simulator units) but no detectable gap under safety. The declared evaluation unit also mattered: under mixed constraints the gap reversed from $+0.374$ to $-0.206$ when weighting shifted from image to lesion-aware cluster. AuthEval thus clarifies whether a study's records support claims about the model, the workflow, or both.

Event pipelines often discretize asynchronous timestamps before learning. This step looks harmless, but its stability depends directly on temporal resolution. We study this dependence at the representation level. We first show that hard temporal binning is discontinuous: an arbitrarily small timestamp shift near a boundary can move unit event mass between bins. We then define a class of nonnegative, mass-preserving, resolution-faithful continuous encoders and prove that every encoder in this class has global L1 sensitivity at least 2/Delta, where Delta denotes bin width. Linear two-bin interpolation attains this limit. Local support and first-moment preservation also make it unique. Experiments on SHD, N-MNIST, and DVS128 Gesture support the analysis. Across uniform timestamp budgets, linear interpolation lowers mean representation drift by 47-72% while keeping clean accuracy nearly unchanged. On DVS Gesture, it produces zero prediction flips across all tested budgets and three seeds. On SHD, measured drift follows 1/Delta with R^2 = 0.992.

Vision-language models (VLMs) and vision-language-action models (VLAs) are increasingly deployed in real-world applications. There, a small perturbation to the recorded camera image may change a decision significantly. However, existing benchmarks for these models only sample perturbations, which does not guarantee the absence of a failure in the untested region. We present the first robustness validation of six VLMs (drawn from the Gemma, InternVL, LLaVA, and Qwen families) and five VLAs (drawn from the GR00T, OpenVLA, and $\pi$ families) over entire continuous regions of photometric and geometric image perturbation: brightness shifts, camera rotations, and their composition. To this end, we build on the validation framework H$^2$V and introduce H$^2$V-M, a margin-aware convergence rule that makes validation affordable at the 32B parameter scale. We demonstrate that H$^2$V-M outperforms H$^2$V by an order of magnitude in model queries and that it finds counterexamples faster than random sampling while providing soundness guarantees. Our VLM and VLA robustness validation shows that robustness is mostly dependent on the perturbation type, rather than the model, and that VLMs are more robust to large camera rotations than VLAs. For VLAs, even perturbations as small as $\pm1^\circ$ can change the commanded action in many cases. We also show that robustness depends more on model family than on model size.

AI agents increasingly orchestrate long-running activities with users, tools, and system resources for days and weeks. To improve agent quality, safety, and cost efficiency, developers need to determine where failures happen, what triggers unsafe effects, and which tasks consume the most budget, then optimize those tasks. In systems software, profiling answers similar questions by aggregating resource consumption and attributing it to responsible code paths to identify hotspots. Yet existing agent observability tools focus on per-execution debugging and tracing rather than cross-run, long term profiling, making these questions difficult to answer at scale. Agent observability needs profiling, not only debugging, but profiling agents is challenging: the responsible entities are task intent like diagnose authentication, compare branches rather than code paths, and lack stable identifiers for aggregation. We propose a semantic operation stack model that adapts profiling to agent trajectories. Uniform operations represent all activities, and operation stacks replace the runtime call stack, enabling hierarchical attribution at different granularities. We observe that an agent's task occupies a contiguous span and decomposes into subtasks, so we introduce recursive operation segmentation, which recursively splits trajectories at task boundaries. AgentPProf is a profiler that aggregates agent trajectories into pprof-compatible profiles, enabling flame graph visualization and analysis. AgentPProf reaches 0.764 $B^3$ F1 against human annotations on CodeTraceBench. On three problem-localization benchmarks, the profile raises MAP by up to 56%, demonstrating that it effectively attributes resources, locates problems, and helps optimize token cost at practical profiling cost. AgentPProf is available at https://github.com/eunomia-bpf/agentsight.

Emphatic temporal-difference learning (ETD) stabilizes the expected off-policy TD update and changes its projection geometry, but neither property determines constant-stepsize sampled dynamics. We construct an ergodic two-state counterexample in which the ETD mean map contracts while the sampled product has a positive top Lyapunov exponent. Regenerative-cycle analysis separates this sign from the infinite variance of the follow-on trace. We introduce regularized emphatic TD (RETD), a normalized first-order post-shock repair that leaves the trace and importance ratios unchanged, stores the emphatic TD signal in a leaky scalar state, and releases a delayed correction. RETD's raw equilibrium is an affine shift of the ETD equilibrium; single- and two-regularization readouts recover the ETD fixed point exactly. We prove almost-sure convergence for harmonic diminishing stepsizes and a conditional constant-stepsize moment-contraction result from a Markovian random-product bound. RETD has certified negative exponents on the two-state construction and one Baird point, whereas the positive Baird ETD sign remains numerical. Paired 10,000-run experiments validate both separations, fixed-point recovery, a nonmonotone stability region, and task dependence. RETD changes post-shock dynamics; it does not reduce the shared follow-on-trace variance.

Simulations based on large language models (LLMs) have proven to be powerful for understanding human behavior, making them valuable additions to the social scientific toolkit. However, LLMs are ultimately black boxes based on deep neural networks which limits their value for social science. This is because of a lack of (i) interpretability: i.e. the ability to assign clear mechanisms driving observed behavior; and a lack of (ii) steerability: i.e. the ability to mute or amplify specific theoretically meaningful mechanisms of action to drive specific model behavior. Here, we demonstrate how the black box could be opened up to further enrich LLM-based simulations. Specifically, we compare three types of methods: (1) prompt-based manipulation, (2) SAE-derived feature steering, and (3) probe-based direction steering and examine their utility for LLM-based social scientific simulations. We do so by interpreting and steering two foundational components of human behaviors, namely preferences (risk attitudes, altruism) and capabilities (divergent creativity, product innovation), operationalized using four classic economic and creative tasks implemented as natural-language interactions. Overall, our results show that SAE- and probe-based techniques often outperform basic prompt-based methods for steering LLM agents, although this advantage depends on the specific prompting strategy involved. Together, SAEs and probes constitute an effective pipeline for social scientists seeking to interpret and steer agents in social simulations: SAEs decompose agents' internal representations into human-readable features, after which probes can reliably shift agents' behaviors in specified directions. We discuss implications of these methods for future work using LLM agents for social scientific simulations.

Pedestrian-count forecasting supports pedestrian-oriented Intelligent Transportation Systems (ITS), including crowd monitoring, pedestrian-traffic staffing and routing, and proactive risk mitigation during surges. Recent time-series foundation models (FMs) report strong zero-shot accuracy on heterogeneous forecasting benchmarks, but it remains unclear whether these gains transfer reliably to pedestrian sensing deployments. We benchmark seven univariate forecasting approaches spanning four paradigms: Seasonal Naive, gradient-boosted trees (LightGBM, CatBoost), deep learning models (N-HiTS, PatchTST), and two pretrained FMs (TimesFM, Chronos-2). Experiments cover two complementary regimes: (i) a five-day special event dataset SAIL2025 at 3-minute resolution with limited in-domain history; and (ii) Melbourne pedestrian sensors as a multi-year hourly dataset (2010–2017) with strong seasonality. We compare the MAE and RMSE results per sensor across datasets and multiple forecast horizons. Results show three consistent findings. First, with limited historical data, Seasonal Naive remains a strong baseline for long-horizon forecasting on high-volume sensors, while trained models can degrade when the next day differs substantially from prior days. Second, boosted trees can be competitive on lower-volume sensors but exhibit higher sensitivity on high-volume sensors under event-driven shift. Third, FMs excel in the seasonal and data-rich regime under long-context configuration. The findings highlight the importance of choosing pedestrian forecasting models based on both the underlying data conditions and the forecasting horizon.

Whitney's theorem allows isomorphism testing for connected simple graphs, apart from $K_3$ and $K_{1,3}$, to be formulated as distinguishing their line graphs. However, the relation between fixed-dimensional Weisfeiler–Leman (WL) expressivity on line graphs and on their roots remains unresolved. We study this relation through Implicit Line-Graph WL (ILG-$k$-WL), which is exactly $k$-WL on $L(G)$, executed over the edges of $G$ with line-graph relations derived from endpoint incidence and without explicitly constructing $L(G)$. On the Whitney-general class, the relation between root-domain and line-graph WL depends on $k$. For $k=1,2$, ILG-$k$-WL adds no distinguishing power beyond root-domain $1$-WL and misses some pairs that $1$-WL separates. For $k=3$, we prove the backward containment $L(G)\equiv_{3\text{-WL}}L(H)\Rightarrow G\equiv_{3\text{-WL}}H$. Strongly regular witness pairs, including the Shrikhande/rook pair, show that ILG-$3$-WL is strictly more expressive than $3$-WL. The backward containment also extends to disconnected graphs with no isolated vertices when every connected component is Whitney-general. Deterministic ILG-$3$-WL separates all three substructure-counting witness pairs, all $105$ pairs in SR25, and $359$ of $400$ BREC pairs. An untrained dense ILG-$3$-GNN gives the same pairwise verdicts on these evaluations.