Science & Mathematics

We investigate multi-constraint policy optimization in constrained Markov decision processes (CMDPs), where interactions among constraint gradients often yield ill-conditioned update directions, causing numerically fragile steps and unintended cancellation or redundant overlap in constraint corrections. To address this, we propose LOCU, which mitigates numerical instability arising from constraint interactions by applying symmetric Lowdin orthogonalization to the natural gradients of the reward and constraints under the Fisher-Rao metric. Combined with Fisher-geometric screening and near-collinear compression, LOCU yields a compact reduced space in which the trust region becomes a Euclidean ball and constraints are handled symmetrically through a low-dimensional active-set solve. For infeasible iterates, LOCU constructs a Pareto-descent direction which preserves the local feasibility of satisfied constraints within the same reduced formulation. Experiments show that LOCU remains stable under high constraint coupling, narrow feasible regions, and infeasible starts, and generalizes across different cost critic architectures.

Large Audio Language Models (LALMs) reason over speech, environmental sounds, music, and other acoustic information. However, their audio encoders produce dense token sequences that make long-form audio costly to process. Existing audio pruning methods reduce this cost using heuristics such as attention, representation similarity, or temporal sampling, but do not explicitly learn which tokens are most useful for a given question. We propose AudioScorer, a lightweight model that scores audio tokens according to the user question. It is trained by distilling the ranking of audio tokens under the frozen LALM's answer-gradient saliency, together with downstream answer supervision. Compared with other pruning methods, AudioScorer achieves the best or tied-best result at 10% token retention on 7 of 10 semantic and acoustic tasks, including speech recognition, emotion recognition, and audio-scene classification.

Near-identical passages can differ in a decisive condition, such as negation. A ranker may order such a pair correctly while placing the preferred passage far down the retrieved list. We examine this distinction using official NevIR pairs jointly covered by fixed candidate pools from dense, learned sparse, and BM25 retrieval. On 2,736 Test query directions, a pairwise-trained ranker improves strict per-query preference over fixed fusion (56.62% versus 52.30%) while reducing preferred-passage mean reciprocal rank (MRR) from 0.5764 to 0.3055. An exploratory fixed-model replay of a variant with background sampling and revised supervision yields MRR of 0.5859 and strict preference of 54.53%. Bounded pointwise reranking of the fusion top 20 improves both outcomes. Qwen3-14B-AWQ with fusion tie-breaking reaches 77.45% strict preference, 58.99% NevIR-style paired accuracy across 1,363 complete pairs, and preferred-passage MRR of 0.7509. A BGE cross-encoder reaches 69.77% strict preference and MRR of 0.7404. A post-hoc analysis finds that the branch where equal Qwen grades for the two designated passages trigger fusion tie-breaking accounts for 27% of evaluated directions and contains most remaining strict-preference failures. These findings support joint evaluation of pairwise preference, designated-passage position, and computational cost, while distinguishing model judgments from the combined ranking procedure.

RNA molecules often function through multiple experimentally observed conformational states. A central challenge is to capture ensemble diversity without sacrificing structural fidelity: samples may collapse toward averaged representative structures or drift into unsupported conformational regions. We introduce REFLEX (RNA Ensemble generation calibrated by FLEXibility), a sequence-conditioned generator for RNA backbone-frame ensembles based on a heteroscedastic stochastic bridge on SE(3). During training, experimental B-factors calibrate residue-wise bridge widths, exposing each observed conformer through local perturbation neighborhoods with residue-specific scales while keeping constrained regions sharper. A learned flexibility condition guides the bridge-field network to recover target conformer-specific frames from these calibrated noisy states. At inference, REFLEX uses only the input sequence, requiring no MSA-subsampling heuristics. On a multi-conformer RNA benchmark, REFLEX achieves the best precision-recall trade-off for ensemble coverage across evaluation thresholds, while recovery-stress ablations indicate gains in target-frame recovery under stronger perturbations.

Autoregressive Diffusion Transformers have emerged as a powerful paradigm for long video generation. However, they are often constrained by prohibitive computational overhead and substantial memory footprints. This inefficiency arises from dense computation mechanisms that allocate uniform processing resources across all spatiotemporal regions, failing to leverage the inherent temporal redundancy of video where motion is typically confined to sparse areas. In this paper, we propose Dynamics-Aware Sparse Attention (DASA), a unified framework that shifts from dense computation to sparse, motion-driven generation to eliminate both storage and computational redundancies. Inspired by video codec principles, our approach explicitly decouples video content into dynamic and static regions. To mitigate memory bottlenecks, we selectively compress the historical KV cache by retaining only dynamic features within the generated chunk. Furthermore, to reduce computational overhead, we employ a region-level computation allocation strategy based on localized dynamics. Additionally, a dynamics-guided distillation method is introduced to enhance performance via a lightweight distillation training process. Experiments on VBench demonstrate that our approach accelerates autoregressive video generation by 2.16x, while preserving high visual fidelity.

When several generators share a fuel supply, one interruption can disable them together and cause electricity shortages lasting days. Such interruptions can be made less frequent, for example by weatherizing gas supply, or shorter, for example by on-site backup fuel. Yet traditional resource adequacy indices, expected unserved energy and loss-of-load hours, fail to distinguish such measures whenever they remove the same expected time without fuel, because for interruptions independent of system conditions these indices depend on the interruptions only through this time. Their difference is revealed by prolonged-shortage exposure, the expected annual hours in shortages of at least a given duration, whose exact chronological calculation must be repeated for every interruption scenario. To reduce this cost, a fast reassessment method with guaranteed accuracy is proposed. Network calculations are performed once and condensed into a restoration response, the additional prolonged-shortage hours caused by one interruption of each duration, from which any scenario is evaluated as a weighted sum without new simulation. An hour-by-hour error bound determines whether this evaluation is accurate enough, and exact dynamic programming is applied otherwise, so that every result lies within a requested tolerance of the exact value on the same Monte Carlo samples. On the Reliability Test System of the Grid Modernization Laboratory Consortium, most evaluations were accepted, fully accepted scenarios typically ran more than 200 times faster than exact calculation, and total computation time decreased in every workload. Measures against fuel interruptions can thus be screened quickly and accurately by their effect on prolonged shortages.

Static type systems for Scheme have traditionally followed one of two paths: manifest type checking with multi-dispatch (Clifton and Leavens, 2005) or latent occurrence typing tailored for untyped interoperation (Tobin-Hochstadt and Felleisen, 2008). In this paper, we propose Polarized Scheme, a static foundation built upon polarized type theory and the duality of computation. In Polarized Scheme, values are positive data producers (A+) and continuations are negative observation consumers (~A+). Functions operate symmetrically under dual viewpoints: as forward value-transformers and as backward continuation-transformers. Computation exists exclusively at the interaction cut executed on a focus-and-context abstract machine separating positive data focus from negative return contexts (R). Delimited control (shift/reset) and continuation marks are expressed directly as Return Stack frame operations parameterized by the prompt's ambient answer type without polluting producer data types. Furthermore, we formalize the hygienic macro system as an interleaved continuation-passing pipeline, where macro expansion erects its own continuation hierarchy nesting into runtime evaluation. We prove Subject Reduction, Progress with respect to prompt-delimited stacks, prove Duality Bisimilarity, characterize Return Stack reclamation for dynamic closures, and report on an executable reference implementation in a verified dialect of Go with a verified cut-elimination dispatcher

Reinforcement learning with verifiable rewards (RLVR) trains a language model by sampling responses, scoring them with an automatic verifier, and updating the policy from the resulting rewards. The policy therefore receives direct learning signals only from trajectories that appear in its rollouts. Yet repeated sampling from the same policy can concentrate on similar reasoning paths, leaving useful alternatives unexplored under a fixed rollout budget. We introduce Activation-Space Exploration (ASE), an online method for diversifying training rollouts. By clustering activation differences between correct and incorrect responses to the same problem, ASE builds a structured bank of latent exploration vectors. Different vectors are then applied to subsequent rollouts to explore alternative reasoning paths. On Qwen3-1.7B-Base and Qwen3-8B-Base trained on DAPO-Math-17k, ASE improves average Mean@8 across six mathematical reasoning benchmarks over GRPO by 4.87 and 1.87 percentage points, respectively, under the same rollout budget.

The rapid evolution of digital circuit complexity demands innovative approaches to hardware design automation. While Large Language Models (LLMs) have influenced software development, their application to Hardware Description Language (HDL) generation remains largely unexplored, particularly for VHDL. This paper presents a comprehensive study for AI-assisted digital circuit design, addressing gaps in current Electronic Design Automation workflows. The work introduces a VHDL dataset of 18,500+ syntax-verified examples from open-source repositories and converted Verilog designs. Five 7B-parameter LLMs were fine-tuned using parameter-efficient QLoRA technique, enabling deployment in resource-constrained environments. To evaluate model efficacy, a rigorous three-level pipeline was established: (1) syntax and synthesis validation, (2) functional simulation with testbenches, and (3) FPGA prototyping on Zynq SoCs. Results show fine-tuned models achieve up to 92% syntax correctness and 73% functional accuracy on common design tasks, with DeepSeek-Coder demonstrating the highest performance gains. Generated designs were successfully deployed on real hardware, confirming not only syntactic but operational viability. This work demonstrates that LLMs, when tailored with domain-specific data and lightweight fine-tuning can contribute meaningfully to the future of generative EDA workflows.

Gene perturbation models aim to predict transcriptional responses to unseen genetic interventions, but comparisons rely on heterogeneous evaluation metrics and differentially-expressed (DE) gene subsets. We investigate the consequences of these choices using eight methods across six single-gene perturbation datasets under a unified evaluation protocol. Varying the number of top DE genes reveals substantial instability in Precision-based rankings, with the winning method changing $86.7\%$ of the time, while rankings based on the rest of the metrics remain comparatively stable. Investigating this discrepancy exposes that repeating a single predicted expression profile across cell-level rows, a standard practice for many methods, can inflate significance estimates and produce excessively large predicted significant-DE sets. Specifically we found that $67.1\%$ of tested perturbation--gene pairs pass the significance threshold despite some of them having minuscule effects. We derive an inflation factor showing that replicated predictions can amplify the standardized Mann--Whitney statistic even after correction for tied ranks, without adding independent information. This artifact changes Precision mainly as the number of DE genes increases because for smaller cutoffs, most of the genes are already significant due to the effect-based ordering. Taking this pitfall into account in our benchmark comparison, we find that a random forest using Gene Ontology features achieves the best overall results.

Relation extraction from aviation accident narratives is challenging because genuine safety relations are sparse among possible concept pairs. This study evaluates a decomposed framework separating relation classification from concept identification. A benchmark of 300 National Transportation Safety Board (NTSB) narratives containing 2,524 unique concept-text nodes was expanded into 21,038 ordered candidate pairs, with only 248 (1.18%) representing RESULTS_IN or CONTRIBUTES_TO relations. A TF-IDF Linear Support Vector Machine (SVM) and weighted RoBERTa classifier were developed using an independent graph-annotated resource with hard-negative sampling and frozen before sealed evaluation. Development Macro-F1 scores of 0.4738 (SVM) and 0.4809 (RoBERTa) declined to 0.2741 and 0.2420 under exhaustive evaluation, with Matthews correlation coefficients of 0.0128 and 0.0197. False positives dominated: the SVM recovered 96 reference relations but generated 7,018 false positives, while RoBERTa recovered 125 but generated 9,824. Post-test diagnostics revealed overlapping RoBERTa confidence distributions and false positives even within valid relation-category directions. Complementary end-to-end SmolLM2 evaluation identified missed and unsupported relations alongside malformed outputs. These findings demonstrate that known concept endpoints do not guarantee reliable relation extraction and that sampled-negative development performance may poorly reflect exhaustive evaluation. Controlling false positives and managing candidate-space sparsity remain essential for aviation safety knowledge extraction.

Electrifying airport ground operations is a key step toward reducing aviation-related emissions, but it makes task assignment and charging coordination strongly coupled in real time. This paper develops a digital twin of airport ground handling operations for electric ground-support vehicles (EGSVs), combining an agent-based model of airport operations with live flight-schedule updates and vehicle-energy dynamics. On this platform, we compare three control strategies for EGSV dispatching and charging: a rule-based heuristic, a rolling-horizon model predictive controller (MPC), and a Monte Carlo tree search (MCTS) controller. The controllers are evaluated on three representative Schiphol Airport operating days and for multiple fleet sizes. The results show a clear trade-off between delays and energy use. Across all tested scenarios, MCTS yields the lowest average flight delay, with a pooled delay of 4.09 min, improving on the heuristic and MPC controllers by 64.7\% and 31.5\%, but at an increased energy consumption of 1\% and 0.3\%, respectively.The MPC and MCTS controllers coordinate charging more effectively throughout the day and reduce the peak charging infrastructure required by 6.0 and 4.7 charger units, respectively. Additionally, we find more vehicles can result in lower overall energy consumption and lower charging infrastructure requirements for all controllers. These findings highlight the value of digital twins not only as offline design evaluation tools, but also as practical environments for real-time control policies in electrified airport operations.

Automatic radiology report generation (RRG) aims to translate medical images into diagnostic text, reducing radiologists' workload and standardizing clinical documentation. Nonetheless, existing approaches mainly focus on single-timepoint analysis and fail to capture temporal disease evolution across longitudinal examinations. While recent longitudinal RRG (LRRG) approaches incorporate historical data, they often combine images from different time points within a single representation space, leading to blurred semantics and inconsistent temporal reasoning. In this work, we propose a Temporal Decoupling with Iterative Mutual-Refinement Model (TIM), a two-stage framework that explicitly decouples spatial pathology from temporal progression and iteratively refines reports through mutual feedback. Stage I performs temporal-decoupled representation learning, separating temporal evolution patterns from disease-specific features and generating radiology reports for both prior and current studies. Stage II introduces a mutual report refinement mechanism that identifies diagnostic inconsistencies within prior reports and iteratively rectifies both prior and current reports through error-sensitive feedback. Experiments on the Longitudinal-MIMIC dataset demonstrate that TIM surpasses existing single-image and longitudinal baselines, achieving new state-of-the-art performance across both language and clinical metrics. Code is available at https://github.com/yihengd/TIM.

On ranked directed acyclic graphs coupled by vertex conflicts we study exact-arrival multi-agent path finding (departure and arrival pinned in place and time) posed as a distributed output task: every cell must emit its own bit of a valid joint path-support mask. The machines are cellular: one automaton per cell, state factored over cells, communication only along a fixed local fabric, synchronous rounds, a query-independent output round, no central decoder, and a per-topology transition code. "Necessary" here means: on an explicit hard topology, over all its feasible queries, no universally correct local-only architecture can renounce either temporal orientation; with global flags, impossibility becomes a bandwidth-round trade-off. The class is not vacuous: a flooding machine solves every instance in as many rounds as the diameter of the local fabric. The bound survives two concessions: we supply every agent's exact individual cavity, which alone solves the single-agent problem in zero rounds and makes the single-agent direction theorem vacuous; and we allow $k$ one-bit global aggregate flags. On a twelve-cell, two-agent, depth-three topology no forward-only machine with $k=0$ is correct on two feasible queries, for any number of rounds; the time-reversed topology does the same to backward-only machines. Placing $p$ copies side by side gives, for forward-only machines, $J_R+kR\ge p$ at constant depth, where $J_R$ measures how much of the query reaches the targets' cones and is $0$ here. A backward-only machine solves that family in constant rounds and a flag-streaming one in $\lceil p/k\rceil+O(1)$, so the bound is tight in order; on the composite family neither orientation can be renounced. The finite lemmas of the gadget and the pair-invariance contradictions are checked in Lean 4; the semantic bridge, the cone lemma and the counting are paper proofs. We report no benchmarks.

This paper provides a counterexample of the sufficient schedulability test for self-suspending real-time tasks under EDF-Like scheduling given by Wang et al. in Theorem 11 of IEEE Transactions on Computers, 74(7):2364-2375, 2025. This paper also shows that their necessary condition is unsound.