Reinforcement Learning (RL) algorithms often rely on data collected from a target environment, which can be expensive and noisy. Although simulators can generate large amounts of additional data, standard simulation mixing approaches, which directly combine simulated and real experience, can degrade performance due to simulator inaccuracies, especially when they bias the solution away from the optimal policy. We propose a conservative simulation-aided learning method based on control variates, where simulated data is used solely to reduce variance in the learning process rather than directly shaping the learning target. We develop our approach for Policy-Evaluation, which is a fundamental building block of RL algorithms. Then we combine it within the actor-critic framework towards policy improvement, as well as extend it towards offline RL scenarios. Experiments on multiple MuJoCo benchmark tasks demonstrate improvements over standard simulation mixing approaches, highlighting control variates as an effective tool for simulation-aided reinforcement learning.
AI is used in much of the clinical trial pipeline. However, selecting the statistical design of a trial, a crucial step that determines how drug doses are tested on humans and evaluated, has seen almost no AI progress. LLMs appear to be a natural fit given their strengths in biomedical knowledge and statistical reasoning. Nevertheless, there is currently no appropriate way to evaluate such abilities because recent regulatory and methodological shifts, most notably FDA's Project Optimus, have redefined what constitutes a reasonable design and rendered many previous designs obsolete. To lay the groundwork for AI-assisted trial design, we introduce TrialDesignBench, a benchmark of 617 Phase-I/I-II oncology dose-finding trials, each paired with an expert-validated reference design that is consistent with the latest FDA regulatory initiatives. This reference standard is produced through a formalization of the expert design process. We evaluate five frontier LLMs, none of which matches the reference standard. Our ablation experiments trace this performance gap to two main causes, inference of latent, unrecorded design determinants from trial sources (supplying them raises class-wise balanced accuracy by 0.22) and structured decision-tree reasoning (+0.24). TrialDesignBench offers a reproducible, contemporary-standard objective for evaluating and enhancing AI-assisted trial design.
Urban traffic flow prediction is a cornerstone of intelligent transportation systems, yet accurate forecasting under adverse weather conditions remains a significant challenge. Existing data-driven models, while effective for routine patterns, often struggle with the sparse data and complex non-linear shifts associated with rare meteorological events. These methods typically treat environmental contexts as shallow numerical features, failing to capture the underlying semantic reasoning-why and how specific adverse conditions alter mobility behaviors across regions with different functional compositions. To address this limitation, we propose Multi-STG, a unified framework that integrates Large Language Model (LLM)-derived semantic priors with graph-based spatiotemporal learning. Multi-STG introduces an LLM-empowered semantic enhancement module that transforms comprehensive context descriptions such as weather, POIs, and holidays into high-level semantic embeddings, effectively providing a knowledge-driven complement to data scarcity. Furthermore, we design a multimodal attentional fusion mechanism to dynamically weigh heterogeneous inputs and a semantic-enhanced prediction head with gated residual calibration to robustly adjust forecasts based on semantic guidance. Extensive experiments on real-world multi-city private car datasets demonstrate that Multi-STG outperforms state-of-the-art baselines, reducing MAE by 11.82\% on average relative to the strongest baseline across three cities, while maintaining stable performance under sparse and non-stationary conditions.
Serving systems choose a weight format for each model and deployment. A low-cost screen predicts the resulting KL divergence from the unquantized model's Fisher information using $\tfrac12\Delta^\top F\Delta$, the expansion underlying Hessian-aware and second-order post-training quantizers. We assess the reliability of this screen in GPU batches with channel-wise int8, per-tensor fp8 and group-wise int4 on Qwen3-8B, Qwen3-4B, OLMo-2-7B, Qwen3-32B and Qwen3-30B-A3B. Relative predictions hold across a 16-fold range of perturbation scales, int4 group sizes and rounding modes, while absolute predictions are heavy-tailed. A trimmed average brings 10 of 12 cells on Qwen3-4B and OLMo-2 within 15% of measured KL. The MoE model exhibits nonquadratic response at small perturbations, with int8 KL scaling as $\alpha^{0.63}$. Computing per-sequence gradients in bf16 can cause fourfold prediction errors. We trace these errors to arithmetic in the lower half of the backward pass and eliminate them by using fp32 there. Batch-invariant kernels run at 0.55 times the throughput, while changes in reduction order introduce a floor of 0.17-0.22 nats per sequence. We build a screener with one streamed fp32 Fisher pass, trimmed tensor-block predictions with a band, a flag for aligned tensors and per-layer format allocation. On the unseen Mistral-7B-v0.3 family, all six predictions fall within the band, all three KL budgets are met, and vLLM throughput is predicted within 0.4%.
Romanized Nepali, Nepali written in Latin script, is common in digital communication but has no standard spelling and few resources, and it is unclear how reliably small open language models can be adapted to it. We adapt three 2-3B base models from different families, Llama-3.2-3B, Qwen2.5-3B, and Gemma-2-2B, with QLoRA on 9,000 instruction examples produced by deterministic transliteration of Nepali instruction data. Each model is trained with one shared recipe and three seeds and evaluated on 1,000 held-out examples with reference-overlap metrics, byte-normalized reference likelihood (bits per byte, BPB), and an output audit, against zero-shot baselines under two prompts. Fine-tuning raises chrF++ from 10-14 to 29.0-30.6 and lowers BPB from 2.23-2.41 to 0.78-0.84; most zero-shot outputs (54-92%) are English-like. Likelihood and overlap disagree on how far apart the models are. Llama-3.2-3B has the lowest BPB in every seed, 0.042 below Qwen2.5-3B (95% CI 0.039-0.046), yet the two cannot be separated on chrF++ (30.51 and 30.61) or BLEU, even with paired item-level tests. Gemma-2-2B, whose tokenizer splits Romanized Nepali words least (2.72 tokens per word), scores worst on every measure. Which model gains most over its zero-shot baseline depends on the zero-shot prompt. In a supplementary comparison with an identical tokenizer, Llama-3.1-8B lowers BPB by 0.053 (0.050-0.056) relative to Llama-3.2-3B, while chrF++ and BLEU rise by only 0.68 and 0.58 points. Lower fertility and higher reference likelihood therefore did not identify the model whose outputs were closest to the references.
Neural avalanches and turbulence-like brain dynamics are usually measured after neural activity has been coarse-grained into continuous fields, leaving open which spike-level mechanisms can generate these macroscopic observables. We build a spatial leaky integrate-and-fire spiking neural network and explicitly transform spikes into activity-density and Hilbert phase fields. A delayed, distance-dependent, passive-dendrite regime reproducibly generates turbulence-like field structure, including high-amplitude phase-defect tracks, structure-function scaling, and power-law-favored continuous-field avalanche tails. Controls show that firing rate, vortex count, and avalanche tail fits are individually insufficient: destroying delayed propagation, dendritic filtering, or spike timing suppresses key field signatures, while random-space nulls can retain many mathematical phase defects but show weaker spatial structure and synaptic-delay transfer. Biologically motivated gates further shape the regime: PV-like perisomatic inhibition clamps phase dynamics, SST-like dendritic inhibition and projection-specific inhibitory facilitation preserve long-lived tracks, and a rate-constrained projection-STP regime expresses a roughly 200-ms episode at 12.25 Hz (full-record mean) across 12 seeds. These results use SNNs as a mechanistic testbed for spike-derived brain turbulence observables and show why phase defects and avalanche-like statistics must be interpreted jointly with spatial scaling and spike-level propagation.
Perceiving objects through fragmented occlusions, such as foliage, is a natural capability of the human visual system but remains a fundamental challenge for computer vision. This work addresses the problem of object reconstruction, or de-occlusion, in the presence of high-density, fragmented occlusions. We propose a reconstruction framework that operates on a sequence of grayscale images of a moving object, acquired by a single camera, either stationary or in motion. The algorithm estimates the object's relative positions across frames by jointly registering the images and optimizing the alignment parameters to maximize the contrast of their average, referred to as the integrated image. This integrated image serves as the reconstructed representation of the object. Experimental evaluations on both synthetic and real video sequences demonstrate the feasibility of accurate reconstruction under occlusion densities exceeding 90%. The proposed approach is competitive with synthetic aperture techniques, which are limited to static objects and require complex camera setup with careful calibration. Furthermore, we show that the integrated image enables reliable object classification under severe occlusion when used as input to standard, off-the-shelf classifiers.
We show that type-logical grammars with the standard rules for product, associativity, extraction and infixation have an NP-complete fixed recognition problem, adapting a construction for first-order linear logic from Pshenitsyn (2025), who already showed that grammars in first-order linear logic are closed under intersection and therefore not semilinear and not mildly context-sensitive. We strengthen these results and shown that grammars in first-order linear logic can simulate the quasi-linear grammars of Book \& Greibach (1970), the grammars recognised in non-deterministic linear time (NSPACE(n) or. NLIN). I argue that NLIN is an interesting class of formal languages as well, and a useful upper bound on the time complexity of formalisms used for natural language, just like LOGCFL is a useful upper bound on the space complexity. Taken together this means that first-order linear logic can be seen as the logic of non-deterministic linear-time string languages, graph languages and automata.
Factor analysis has been central to the analysis of high-dimensional biomedical data. However, existing exploratory and confirmatory factor analysis methods face several limitations: less interpretable factor findings, difficulty in identifying which variables can and cannot be represented by low-dimensional factors, challenges in estimating the covariance among factors, and maximum-likelihood fits that break down when variables outnumber subjects. We propose graph-guided Bayesian factor analysis (gg-BFA) to address these gaps. We first use a Bayesian model to explore and learn the latent graph structure, identifying which variables can and cannot be represented by low-dimensional factors and how many factors are needed. A dense-subgraph search proposes partitions of the variables into communities, each defining one factor, and a forward stepwise search compares their fractional variational posteriors and stops when the evidence lower bound first decreases. We show that this search recovers the true communities even when the number of variables grows much faster than the sample size. We then develop a confirmatory Bayesian model whose variational posterior of the covariance contracts at the parametric rate up to a logarithmic factor, yielding posterior inference for the factors and their covariance, as well as predictions of missing values. Its four members range from a community mask to several cross-loadings per variable. In five biomedical panels, the complete analysis took under ten minutes each on one GPU, and cross-loadings improved held-out covariance and the prediction of masked measurements. In TCGA Kidney, six of the first eight factors were enriched for known pathways, and 190 of the 210 factor correlations had 95% credible intervals excluding zero. Overall, gg-BFA is computationally fast and provides interpretable factor analysis findings and reliable inference results.
AI safety evaluations often involve sensitive prompts and proprietary models. Evaluators want to keep their evaluation prompts secret from model owners to prevent purposeful or incidental data leakage, ensuring longer usefulness of their benchmark. Model owners in turn wish to keep their proprietary model weights and inference code within their own secure infrastructure to protect their IP. We present a way forward for "double-blind evals", which provide cryptographic guarantees that an evaluator's prompts are not shared with the model owner, while simultaneously maintaining the confidentiality of the model weights and inference code. To accomplish this, we use secure enclaves that encrypt data in RAM at the hardware level. In a world-first pilot, we test this approach by evaluating a proprietary Google DeepMind model (Gemini 2.5 Flash Lite) against a selection of the private AILuminate benchmark corpus, using an NVIDIA H100 secure enclave on Google Cloud Platform, and using OpenMined's PySyft to maintain data and model privacy. We believe this effort marks the first use of this technology to provide mutual secrecy for one organization's frontier-class proprietary model weights and another organization's closed benchmark.
Reconstructing audio from brain activity remains challenging. Existing brain-to-audio pipelines typically map neural responses to predefined stimulus representations before synthesis. While recent work in the visual domain has explored direct conditioning on brain activity, this approach has not yet been extended to audio diffusion models. We introduce MAAS (Multi-Area Auditory Synthesis), an end-to-end framework that conditions a pretrained audio diffusion transformer on ROI-wise fMRI responses without first supervising a decoder to predict an external stimulus representation. MAAS uses region-specific projections and learned brain tokens to condition Stable Audio through a dedicated cross-attention pathway. On a 7T fMRI dataset, we demonstrate consistent improvements over baselines, including two-stage fMRI-to-latent conditioning, across both acoustic and semantic evaluation metrics. In addition, we introduce a fidelity metric grounded in a computational model of auditory neural processing, which allows us to quantify the contribution of individual auditory cortical regions to reconstructed audio representations.
The traveling thief problem (TTP) asks a thief to choose a tour and a packing together, and the benchmark of 9,720 instances of Polyakovskiy et al. (2014) is the standard test set for comparing TTP heuristics. We show that four of these instances are not TTP instances at all. Their files give a renting rate of 0.00, so the tour has no effect on the objective and each instance reduces to a 0-1 knapsack problem. One of them belongs to the 60 categorized instances on which new methods are usually compared, and on it five of nine published methods report exactly the same value. The cause is the format of the files, which keep the renting rate with two decimals, while the rule of the generator gives a rate close to 0.003 on these instances. The same rounding may shift the rate of 949 other instances by more than 1%. Since the tour and the packing used by the generator were not published, the original rates cannot be recovered. We therefore define new rates with the same rule, applied to a fully specified reference solution that we publish, and we show with a published heuristic that the corrected instances behave as TTP instances again. We also give reference results on the corrected instances and recommendations for future studies. Finally, we present ttplib, a new open-source Python library that loads any of the 9,720 instances, including the corrected ones, without downloading the whole benchmark.
Large language models (LLMs) are increasingly used to translate technical explainable AI (XAI) artefacts such as SHAP into explanatory narratives (XAINs) for lay users. In high-risk domains, these explanations are often the main resource through which individuals can understand and contest adverse decisions. Yet LLMs are known to reproduce social bias, even from proxy variables alone, and fairness in XAINs remains unstudied. Because XAINs are bound to a ground-truth artefact, fairness concerns not only what a narrative says and how, but also whether it is equally faithful across social groups. We introduce fairness in XAINs as a new field of research and propose a framework that generates narratives under three conditions (protected attribute disclosed, withheld, or counterfactually swapped), allowing five comparisons that pinpoint the cause of disparities. We apply it to adversely classified instances from four datasets in credit scoring, education and employment, generating narratives with five LLMs, and evaluate faithfulness, feature coverage and sentiment. We find significant disparities across all datasets, metric families and models, arising in two ways. Faithfulness disparities result from feature values: LLMs make plausible but false statements, triggered by values that are unequally distributed across social groups. Feature coverage and sentiment disparities stem from the protected attribute itself: once it is disclosed, LLMs give less explanation to, or frame more negatively the narratives of, groups that have historically been discriminated against or are stereotypically associated with worse outcomes. A more faithful model removes the faithfulness disparities, and withholding protected attributes mostly removes the others, but neither suffices on its own. Fairness in XAINs should therefore be audited separately from the predictive model and its XAI artefact.
Optimizing molecules under small evaluation budgets requires deciding both where in chemical space to search and which candidate within that region to evaluate. We introduce Trust Region Adaptive Chemical Exploration (TRACE), a framework for open-ended molecular Bayesian optimization (BO), where candidates are generated on demand rather than selected from a fixed library. TRACE combines an objective-free graph-perturbation mechanism with a fixed, chemistry-aware Gaussian process (GP) surrogate using fragprints, a representation combining molecular fingerprints and fragment descriptors. Rather than learning the search geometry itself, TRACE makes the scale of chemical exploration explicit through a Tanimoto-similarity threshold controlling chemical extent and a cardinality budget controlling candidate-search breadth. A simple success/failure controller expands or contracts both quantities during optimization, while the GP and acquisition function select which admissible molecules to evaluate. Across six challenging molecular optimization benchmarks under matched 1,000-evaluation budgets, TRACE achieves the strongest overall sample efficiency as measured by the best molecule found as evaluations accumulate. Crossed ablations on two objectives further show that adaptive region control contributes beyond the proposal mechanism, while its benefit depends on the proposal distribution. We further apply TRACE to the stability of iron phthalocyanine (FePc), a non-precious molecular electrocatalyst whose degradation can involve loss of the Fe center. Using a first-principles density functional theory (DFT) workflow designed for accurate energy evaluation, TRACE increases the computed thermodynamic resistance to Fe dissolution from 0.42 to 2.59 eV over 80 designed evaluations.
Reinforcement learning with verifiable rewards ordinarily spreads a single advantage over a whole trajectory, a granularity too coarse to isolate the few decisions on which success in a long, multi-turn interaction actually turns. Privileged self-distillation supplies denser supervision, but it leaves unanswered how a set of local teacher-student discrepancies should be converted into credit along a sequence. We introduce AgentOPSD, a critic-free scheme that assigns credit recursively, one interaction turn at a time. Inside a turn, the token-level log-probability gaps between a skill-conditioned self-teacher and the student are summed, and that sum acts as evidence in a recursive Bayesian update carried out in log-odds space; the resulting revisions reallocate the outcome-derived advantage across turns through a multiplier that is bounded and preserves its sign. With Qwen2.5-3B and Qwen2.5-7B as policy backbones on ALFWorld, the method improves consistently on GRPO and on self-distillation baselines, and attains 89.1% success at the 7B scale. Ablations attribute the gain to aggregating evidence at turn boundaries and to making the belief revision depend on history, rather than to the local gap by itself.
Likelihood-based methods are widely used to learn the parameters of partially observable Markov decision processes (POMDPs). While finite-sample guarantees for policy suboptimality exist for certain POMDP subclasses, analogous guarantees for parameter recovery remain less well understood. For any candidate POMDP parameter, we bound its parameter recovery error in terms of its likelihood gap to the ground truth, the structural conditioning of the observation and transition kernels, and the behavior policy used to collect data. The bound applies in both the undercomplete and overcomplete regimes, with the latter following naturally from the undercomplete analysis. Because the result applies to any sufficiently good candidate, it yields finite-sample guarantees for the maximum-likelihood estimator and any procedure producing a parameter with controlled likelihood gap. This is particularly useful for near-optimal learning in infinite-horizon POMDPs, where existing approaches largely rely on model-based planning with accurate transition and observation kernel estimates. The bound separates the contributions of the behavior policy and model, showing how data-collection choices, including action coverage, and their effects on mixing can accelerate parameter learning.
As vision-language models (VLMs) move toward interpreting medical images directly, their outputs increasingly include not just a label but a stated confidence and a written justification. Yet the field still predominantly evaluates VLMs using just accuracy. We evaluate the coherence between answer, confidence, and explanation across four VLMs on breast density and BI-RADS categorization, using two mammography datasets. We perform controlled perturbations of the question, multiple choice option orders, and the images and assess changes in the three outputs. Our results suggest confidence does not decrease even as the mammography image is changed. Shuffling the options also changes most VLMs' answers. Notably, there are cases where a model's own stated reasoning contradicts its final answer, a distinction accuracy alone cannot capture but one that governs whether an AI-generated answer can be trusted.
Many tasks compose transformations by a known rule: chained rotations in robotics and navigation. Learned models relearn it, compounding error; classical estimators hardcode it and miss hardware deviations. We impose the known structure and learn only the remainder: an exact non-commutative group law sits in the node of a self-contained layer, its edges learn local increments, and a zero-initialized correction psi learns unknown deviations of the law. A pre-training criterion predicts exact representability. On fixed-angle SU(2) composition, a 54-parameter dense edge is the most accurate dense model, at 27x fewer parameters than the strongest fixed-matrix baseline, and lies 0.05 deg above the Bayes floor. On real robot arms, an isolated 306-parameter psi reaches 1.12-1.14 mm versus 1.29-1.33 mm for Gauss-Newton at N >= 60, and the recipe transfers across three vendors and two protocols. The recipe matches classical fits where the data obey the law and loses where the imposed law is not the task's.
Vision language models (VLMs) encode substantial expert knowledge within their weights, yet their direct application to specialised classification domains such as benthic imagery remains limited. Meanwhile, vision transformers trained on these tasks rely predominantly on low-level visual latents (colours, textures, and shapes) that lack the higher-level reasoning humans employ for identification. We propose Guided Hierarchical Multi-Label (GHiML) learning, a pipeline that extracts conceptual latent variables from a VLM conditioned on training data, generates sample-relevant captions grounded in those variables, and fuses the resulting textual embeddings with image embeddings via cross-attention and a per-node gated router. We also introduce the inverted HML score ($d_{\mathrm{inv}}$), a similarity score that evaluates hierarchical multi-label predictions through graph traversal distance and a cardinality penalty, providing an interpretable alternative to existing measures, which are insensitive to realistic degradations in sparse, deep hierarchies. Experiments on BenthicNet, FathomNet and iNaturalist show that the router improves macro $F_1$ over the vision-only baseline by up to 6.7 pp and $d_{\mathrm{inv}}$ by up to 21.1 pp on in-domain benthic data, with gains persisting under distribution shift. Ablations locate the contribution of the text itself: on the multi-label biotic hierarchies, removing the per-sample text signal degrades $d_{\mathrm{inv}}$ while leaving macro $F_1$ unchanged, consistent with text acting through prediction cardinality rather than node discrimination, whereas iNaturalist, whose annotations are single-label, loses 18.7% of macro $F_1$. Our results support the hypothesis that VLMs contain exploitable specialised knowledge that, when appropriately elicited, can augment classification in domains where direct visual learning is insufficient.
Video world models conditioned on an input video make it possible to explore real, recorded dynamic scenes, enabling re-shooting of footage, free-viewpoint replay, and real-to-sim environments for embodied agents. Such uses require the generated rollout to stay faithful to the recorded world, yet existing benchmarks start from a single image or a text prompt, and video-conditioned generation is evaluated mainly on perceptual quality and camera following. These evaluations do not directly assess whether exploration preserves subject identity, event progression, and scene consistency upon revisiting a location. We introduce WanderBench, a benchmark for exploring dynamic worlds from video. It comprises 745 single-shot source videos of real and AI-generated content in three difficulty tiers (3-25 s), paired with camera routes given as keyboard actions, and evaluates six dimensions with 24 human-validated sub-metrics. Benchmarking eight models, we find that as clips grow longer, models keep the static scene and visual quality but lose camera control and the source dynamics, and that none keeps revisited places consistent. We further propose WanderLand, a training-free framework that conditions each frame of a frozen model on the same-instant source and, on revisits, on its own earlier static content. WanderLand improves five models on every dimension, by up to 8.3 points in source-dynamics fidelity.