Papers

On-policy distillation (OPD) provides teacher supervision on student-generated trajectories, but a single trajectory realizes only one continuation at each decision, limiting supervision coverage under a finite sampling budget. We propose BranchOPD (Branching On-Policy Distillation), which selects informative decision points using student uncertainty and teacher-student disagreement, resamples student-policy continuations from shared prefixes, and applies token-level teacher supervision to both main trajectories and branches. BranchOPD operates at token-level decisions in mathematics and within-turn decisions in agent tasks. Shared-prefix reuse reduces student-generated token costs by approximately 19% on ALFWorld and 31% on WebShop relative to seven independent trajectories. We also introduce protocol-boundary repair (PBR), which uses targeted supervision under protocol constraints to mitigate repetition and wasted generation. BranchOPD outperforms Vanilla OPD and Multi-Rollout OPD (n=7) across all evaluated benchmarks, with gains over Vanilla OPD of 18.55 percentage points in mean mathematics accuracy over 16 sampled answers (avg@16) and 13.8 points in mean ALFWorld ID/OOD success. Mechanistic analyses and selection ablations show that branching at uncertain decisions expands subsequent behavior coverage and outperforms random selection and turn-initial branching.

Nonlinear model predictive control (NMPC) can be applied to large nonlinear systems with complex objectives while enforcing constraints, but the cost of solving the underlying optimization problems limits real-time application for systems with fast dynamics or many states. Neural networks can reduce this online cost by approximating the control law, but may violate constraints and typically require many precomputed optimizer solutions for training. The learning-based iterative solver for constrained optimization (LISCO) instead learns to predict primal-dual steps toward satisfaction of the Karush-Kuhn-Tucker (KKT) conditions through self-supervised training without precomputed solutions, but its dense, monolithic architecture grows with the prediction horizon. We remove this horizon dependence by exploiting the temporal structure of the problem in the solver architecture, which makes the network size independent of the horizon while leaving the training loss and the solver iteration unchanged. A closed-loop resampling strategy further aligns the training distribution with closed-loop operation at no additional sampling cost. On an economic NMPC benchmark of a CSTR cascade, the proposed solver scales to larger problems than monolithic LISCO and achieves median speedups of up to a factor of four over IPOPT. In closed loop, it achieves comparable control performance to IPOPT with no observed constraint violations.

Generating edge case scenarios is crucial for training, testing and safety assuring autonomous systems. This identifies hazardous failures to increase system reliability and robustness, ensures the system can still function safely on the boundaries of its operational design domain and learns the expected system behaviour for these edge cases. The rapid advancement in the generation and analysis capabilities of language models (LMs) has expanded their applicability and recent work has investigated LM scenario generation. However, many recent approaches rely on human-in-the-loop or rule-based strategies. This restricts the LM's ability to generate diverse and edge case scenarios; the long tail of the distribution. Our work uses local LMs in a laptop-based agentic AI framework for adaptive scenario generation. This is cheap, fully-controlled, ensures data privacy and security. It uses think, act, observe, update, and continue loops with lightweight JSON encoding to generate scenarios. We use the system requirements, system description and environment description as knowledge bases in contextual prompts for an LM to ensure scenario realism. A second, independent LM critiques each scenario using domain specific criteria so it can be refined. The refined scenarios are run in simulation and evaluation metrics collected. These metric scores are analysed and similar scenarios are created for edge cases (exploitation) or novel scenarios are generated to explore the scenario space. Post-hoc clustering using metric scores and thematic-type analysis of the rich-text scenario descriptions produced by the LM pinpoints functional gaps and their triggering conditions. This creates traceable evidence for the safety assurance case. Our work highlights the potential of a flexible and adaptive local agentic AI framework for automating edge case generation and producing traceable safety assurance evidence.

We derive approximation rates for Sobolev functions defined on the complex unit sphere and complex projective space, representing quantum states, and on the special and projective unitary groups, representing quantum unitary operations. By using harmonic analysis on compact groups and compact homogeneous spaces, we characterize the corresponding Peter–Weyl decompositions and Sobolev spaces via irreducible representations and the Laplace–Beltrami operator. Combining spectral truncation with finite-depth quantum-circuit approximation, invoking the Solovay–Kitaev theorem, we establish quantitative rates that separate the spectral truncation error from the circuit approximation error.

Representational similarity measures (RSA, CKA, Procrustes) are chosen for the transformations they treat as irrelevant, and all of them take the set of conditions as given. That set is fixed by two decisions, which inputs are admitted and how they are grouped, both taken before the matrix a measure operates on exists: no invariance property reaches them. We measure what they are worth where the arrangement the conditions ought to have is known independently of the systems compared, the acoustic organisation of vowels, across fifteen languages and twenty-seven representations. Changing which categories are compared, at a fixed number of them, moves the gap between two representations by more than twice the effect being measured, while changing how many moves it not at all. In this setting rankings of models are identified under these choices, and rankings across languages are not.

Modern attention workloads require diverse masking patterns whose boundaries depend on runtime metadata, yet efficient attention kernels natively support only a small set of patterns. Existing methods for supporting custom masks in these kernels face an expressiveness-efficiency trade-off. Predicate-based methods are expressive but incur substantial preprocessing and execution overhead, whereas interval-based methods execute efficiently but restrict expressiveness through a predefined interval count. We present CompileMask, which addresses this trade-off by connecting expressive mask specifications with efficient attention execution. We design a compiler that analyzes visibility rules during the offline stage and generates a mask-specific preprocess kernel. We further propose MaskSpec, a region-based DSL that lets users express these rules. During the online stage, the generated kernel reads batch metadata and produces a compact Execution Descriptor that lets the attention kernel efficiently check visibility by comparing element indices against precomputed endpoints. Experiments show that CompileMask achieves up to $3.8\times$ speedup over state-of-the-art methods.

We consider the reachability problem in fully dynamic directed planar graphs: the graph undergoes edge updates (insertions and deletions) and we receive queries asking if one vertex is reachable from another. The current-best online algorithm for dynamic planar reachability spends worst-case time $\widetilde{{O}}(n^{2/3})$ per query/update [Subramanian, ESA '93]. This can be improved to $\widetilde{{O}}(n^{1/2})$ in the offline setting or when the embedding is fixed [Diks and Sankowski, ESA '07]. Our contribution is twofold: 1. We present the first algorithm that breaks the square-root barrier for dynamic planar reachability in the offline setting, obtaining time ${O}(n^{0.49})$ per query/update. This shows that the problem is easier than answering distance queries in a dynamic undirected planar graph, for which a conditional lower bound of $n^{1/2 - o(1)}$ is known [Abboud and Dahlgaard, FOCS '16]. Notably, this lower bound applies to both online and offline algorithms. 2. Second, in the general online setting we design a randomized data structure with amortized update and query time ${O}(n^{0.664})$. This is the first improvement upon the exponent $2/3$ in over three decades. The main engine behind our theorems is a new sparsification framework aimed at maintaining information about reachability. Specifically, we study digraphs through the lens of VC dimensionality of certain set systems. First, we design sparse reachability emulators for planar digraphs that preserve the VC dimension. Second, we utilize orderings with a low “stabbing number” [Chazelle and Welzl, Discrete Comput. Geom. '89] to compute subquadratic reachability emulators in subquadratic time whenever the VC dimension is bounded.

Metaheuristic comparisons are often reported at evaluation budgets far larger than an expensive engineering simulation can afford. This paper presents REMBGO, an evaluation-budget-aware refinement of EMBGO. Its contribution is not any single operator: it is the integration of population preservation, scale-normalized exploration and recovery mechanisms whose parent-relative benefit is concentrated under restricted function-evaluation budgets. REMBGO is evaluated on CEC2022 at a primary, application-aligned budget of 1040 function evaluations, with a surrounding budget ladder reported descriptively to characterise budget dependence rather than as a family of controlled hypotheses. A component-removal study finds no individual mechanism whose removal survives multiplicity correction, consistent with an integrated rather than a single-operator contribution. The algorithm is then applied to EAF–STATCOM controller tuning, and the resulting frozen gain vectors are transferred without retuning to twenty pre-specified unseen stochastic arc realizations ($210$ fixed-gain simulations, all feasible). The selected gain-vector main effect accounts for only $3.6$–$5.8\%$ of the crossed validation variation, against approximately $61.4\%$ for interaction plus residual, and the matched REMBGO–EMBGO paired differences are controller-dependent in sign and not statistically significant. REMBGO is therefore presented as a resource-aware refinement whose benefit must be stated together with the evaluation budget, not as a universally superior optimizer.

Video reward optimization can increase proxy rewards while reducing video dynamics, a mismatch we identify as motion collapse, including under a video-level motion-quality reward. We propose GraftOOD, which replaces part of the online rollouts in selected prompt groups with out-of-distribution video exemplars. A shared velocity-regression objective combines group-relative reward rectification with sequence-level exemplar supervision, requiring neither exemplar generation trajectories nor an additional supervised-loss coefficient. We characterize its pointwise optimal field as a source-posterior-weighted combination of the two source-specific target fields. We construct GraftSet-Video, containing 9,588 format-matched exemplars for 2,397 prompts, to supply external supervision. On Wan2.1-1.3B, GraftOOD raises Dynamic Degree from 4.86 to 44.41 relative to reward-only FlowAWR, while HPSv3 decreases from 8.10 to 7.32 and VisionReward remains close. At the main setting, GraftOOD leads nine of ten metrics among three exemplar-augmented methods. Ablations of exemplar frequency and within-group count reveal dynamics–reward trade-offs rather than uniform gains from increasing exemplar usage. These results support external exemplar supervision as a complement to scalar reward feedback, without equating higher measured dynamics with coherent, visually plausible motion in the generated videos. Code is available at \href{https://github.com/Fuzm7/FlowAWR}{GraftOOD}.

Overtourism concentrates visitor demand on few iconic sites, while other points of interest (POIs) in the same area often remain known only to a narrower public. This imbalance is especially critical in mountain environments, where limited infrastructure amplifies environmental pressures. Discovering lesser-known POIs is relevant for redistributing tourism demand spatially, valuing underrepresented heritage, and temporally, easing high seasonality pressures on popular POIs. Across the literature, definitions of lesser-known POIs or hidden gems remain scattered, and existing scoring approaches remain incomplete, relying on narrow indicators (e.g., number of photos, reviews, likes) which fail to characterize mountain POIs holistically. These approaches are further constrained by their reliance on commercial proprietary sources, or by questionnaires that are subject to memory bias. To address these gaps, we introduce a novel multi-dimensional metric, the Obscurity Score, that groups fragmented literature into four key dimensions (International Awareness, Attractiveness, Promotion, and Accessibility) to systematically detect and characterize lesser-known POIs and enable their promotion as a first step toward mitigating overtourism. To operationalize this score, we develop a generic pipeline mapping open data sources onto these four dimensions. Preliminary results from a web-based prototype demonstrate the practical relevance of the proposed approach.

CAD segmentation datasets are predominantly synthetic or automatically labelled, leaving it unclear how well existing methods transfer to real-world engineering tasks. Fastener-relevant hole detection represents one such task, requiring human annotation and exhibiting greater geometric complexity than commonly used benchmarks. We address this gap through three main contributions. First, we develop two task-specific rule-based methods, one graph-wire-search and one multi-view based, and complement them with a human-annotated benchmark dataset derived from engineering CAD data that also enables training of UV-Net and AAGNet, representative learning-based approaches for B-Rep segmentation. Second, we conduct an extensive experimental study including parameter optimisation, evaluation on our benchmark dataset, cross-domain generalisation on a synthetic dataset, and an assessment of the feasibility of the approaches at varying dataset sizes and IoU thresholds. Third, we provide evidence-based insights into the strengths and limitations of different approaches: rule-based methods generalise well without training data, with the graph-wire-search method favouring recall and the multi-view method favouring precision, whereas learning-based methods achieve strong performance with sufficient annotations but exhibit reduced robustness under domain shift. These findings highlight trade-offs between data requirements, and robustness, and provide guidance not only for hole detection, but more broadly for real-world CAD segmentation tasks.

Multimodal LLMs can fail to read text embedded in an image's coarse structure (text a human recovers by squinting) and report that no text is present. We name this class Squint Text and study it as a recognition deficiency, motivated by cases from a production moderation pipeline. We introduce SquintBench, comprising 1076 production-derived moderation images, and SquintCtrl, a controlled synthetic suite for mechanism diagnosis. Frozen-weight analysis shows that hidden text leaves a coarse signal while glyph identity is weakly represented; identity becomes substantially more accessible when the correct spatial regions are supplied, implicating spatial selection as a major bottleneck. Suppressing high-frequency content also improves reading at a fixed visual-token budget, while visual token count may also contribute. These findings motivate Squint OPSD: a frozen teacher receives a downscaled view and the ground-truth transcription as privileged context, and its token-level distributions supervise a student along on-policy trajectories. At inference, the student reads the original image alone. Squint OPSD reaches 88.61 ANLS on SquintBench, versus 26.50 for the unadapted backbone, 24.44 for the best general-purpose model on the original view, and 68.60 at best for an off-the-shelf model under a downscaled view. These results substantially narrow the gap between available visual evidence and its generated read-out without auxiliary views at inference.

The BROJA partial information decomposition provides an operationally motivated way to separate the information that two sources carry about a target into unique, redundant, and synergistic components. For continuous and high-dimensional variables, however, computing the BROJA decomposition requires solving a difficult optimization problem. Previous work showed that, when each source-target pair is jointly Gaussian, the problem can be reduced to finite-dimensional numerical optimization. Here, we derive a closed-form solution to this optimization problem, enabling direct computation of the BROJA decomposition in the general multivariate Gaussian setting. The resulting closed-form expression enables speedups exceeding two orders of magnitude for high-dimensional Gaussian BROJA PID. Code is available at https://github.com/brainsandmachines/Eigen_PID.

We prove that the cyclic sum $\sum_{i=0}^{10}(x_i-x_{i+1})/(x_{i+1}+x_{i+2})$ is nonnegative for every strictly positive real eleven-tuple, with indices taken modulo eleven. Clearing denominators reduces the problem to a homogeneous polynomial. Subtracting the smallest coordinate then separates the proof into nonnegativity of a common-translation derivative and nonnegativity on a full boundary face. We supply exact rational polynomial certificates in Gram-matrix form for both statements. The derivative certificate uses a linear multiplier, centered monomials and cyclic averaging; the boundary certificate uses a strictly positive quadratic multiplier and uncentered monomials. Their verification requires 6,069 cyclic coefficient comparisons and 34,304 literal coefficient comparisons. All 606 certificate matrices are certified by integer congruences with strictly diagonally dominant images. A Lean 4 development formally verifies the certificate checks and their connection to the original inequality. The exact certificate data, replay programs, and formal sources are publicly available.

Author Name Disambiguation benchmarks are predominantly derived from closed infrastructures that may not reflect conditions in open bibliographic environments. We present OC-AND, an open citation-enriched benchmark for AND constructed by remapping the WhoIsWho dataset with bibliographic and citation metadata from OpenCitations. The dataset is produced through a reproducible five-stage pipeline involving DOI verification, metadata validation, OpenCitations enrichment, and author-centric transformation. OC-AND retains the WhoIsWho ground-truth files and partition structure unchanged, while its publication files are restricted to publications that could be matched to a validated DOI and an OpenCitations record. It introduces realistic characteristics of open environments: heterogeneous metadata completeness, absent abstracts and affiliations, asymmetric citation coverage, and sparse graph connectivity. The dataset contains author records with associated citations, enabling evaluation of AND methods in settings closer to real-world open scholarly infrastructure scenarios. OC-AND is released as open data with complete provenance documentation and pipeline code, supporting reproducible research and the development of citation-aware disambiguation approaches.

AI systems are making rapid progress in mathematics, but this progress is hard to measure. Benchmarks of solved problems saturate, and in benchmarks of open problems each problem can be resolved for the first time only once. We present SqueezeBench, a benchmark that is resilient to saturation: a success raises the bar instead of removing the problem. Each problem fixes an integer function $f(n)$ whose values are known only for small $n$, and the benchmark keeps a lower and an upper record on it. A record is a computable Lean function with a machine-checked proof that it bounds $f$. A submission proposes a new function and proves that it bounds $f$, that it is no worse than the current record at any $n$, and that it is strictly better at some $n$. Once accepted, it becomes the record that the next submission has to beat. The system thus chooses the bound as well as its proof, and it may improve the record at a single $n$ or its growth rate. At every $n$, the gap between the two records shows how far the certified bounds are from determining $f(n)$. Every proof is checked by two independent Lean kernels, so no referee needs to check the mathematics. The checks also reject a function that computes $f$ by exhaustive search at small $n$, and a maintainer only reads the submitter's report and looks for such a search at larger $n$. We release ten problems from extremal combinatorics with twenty proved initial records and the admission tools. As a baseline, agents based on a frontier language model, with at most one hour per record, had 82 submissions accepted and improved every record, without going beyond the published literature. Agents based on two newer models then started from these records, had 60 more submissions accepted and improved 19 of the 20 records again. For eight of the ten problems, at least one record is still asymptotically behind the best published bound.

Cyber-Physical Systems are developed through multiple model based activities, comprising several stages of development. As the development progresses through simulation, experimentation, and operation, observations of system execution may reveal missing, uncertain, or invalid design knowledge. The path from execution results to persistent design-time enrichment, however, is often implicit. This paper addresses this challenge by proposing an evidence driven workflow for systematically transferring knowledge produced during simulation and operation back into design-time models. We demonstrate the workflow using an Influence Model, a design-time artefact used to represent how design artefacts and environmental factors jointly affect system response properties, using a sign-following robot use case. We show how simulation evidence can enrich Influence Models by supporting participant pruning decisions, interaction characterisation, influence-function abstraction, and validity-scope information. The application demonstrates the feasibility of the workflow for simulation-based evidence and its potential benefits to improve design-time reasoning and feedback in subsequent development iterations.

Large language models (LLMs) have taken the world by storm. Their ability to communicate with us in our own language, and to reason about our world, has made them a favorite tool for writing, answering questions, and much more. Their answers already influence our decisions and our actions. What would happen if LLMs themselves could act in the world directly? What would happen if LLMs could interact with each other? Agentic LLMs combine the power of large language models with the ability to act in the real world. In this book we will discuss how agentic LLMs work, how they act, how they interact, and what this means for us, as users.

Reinforcement Learning (RL) substantially improves planning performance in autonomous-driving Vision-Language-Action (VLA) models. However, existing RL methods rely on scalar trajectory rewards that measure overall quality but fail to identify specific planning errors or provide actionable correction signals. In long-tail scenarios, errors in scene understanding, behavior selection, and trajectory execution are compressed into a single score. We observe that Reflective Guidance turns trajectory evaluation into explicit diagnoses and corrective instructions, substantially improving planning accuracy. Yet such guidance requires an initial prediction and its evaluation results, making it unavailable for direct single-pass inference. We propose Reflect-VLA, which treats Reflective Guidance as training-time privileged information and internalizes its planning benefits through Dual-Path On-Policy Self-Distillation (Dual-OPSD). Unlike scalar-reward optimization, Dual-OPSD uses the driving score only to route student-generated trajectories. It distills dense token-level corrective distributions from the teacher for risky trajectories while applying self-imitation to preserve reliable ones. To provide reliable corrective distributions on student rollouts, the teacher is progressively constructed through Driving-Knowledge Pretraining, Planning-and-Refinement SFT, and Self-Refinement Rejection Sampling. At inference time, Reflect-VLA generates trajectories in a single pass without explicit guidance or iterative refinement. Experiments on NAVSIM v1/v2 and Bench2Drive demonstrate competitive planning performance while reducing training time by over 90% compared with RL.

We give deterministic algorithms for approximating the girth~$g$ of an undirected graph with $n$ vertices, $m$ edges, and nonnegative real edge lengths. For every integer $k\geq2$, our algorithm returns a cycle of length at most $\frac{2k}{3}g$ in $O((m+n^{1+2/k})k\log n)$ worst-case time. For every constant~$k$, this matches the stretch and the expected running time of the randomized algorithms of Kadria, Roditty, Sidford, Vassilevska Williams, and Zwick [SODA'23, SICOMP'26] and Kadria, Roditty, and Vassilevska Williams [ESA'26]. In general, it loses only a factor of $O(\min\{k,\log n\})$ in the term that depends on~$m$. The best previous deterministic trade-off, of Kadria, Roditty, Sidford, Vassilevska Williams, and Zwick [SODA'22], has stretch~$k$ instead of $\frac{2k}{3}$ at the same exponent (for even~$k$) and requires integer edge lengths. In particular, for $k=3$ we obtain a deterministic $2$-approximation in $O((m+n^{5/3})\log n)$ time for real edge lengths, whereas the deterministic algorithm of Ducoffe [SIDMA'21] for real lengths has stretch $2+\varepsilon$. Our main technical idea is that a deterministic construction of the Thorup–Zwick hierarchy used by these algorithms may stop exploring as soon as it finds a cycle: the approximation analysis only needs clusters truncated at half the best cycle certificate found, so balls stopped in this way need not be hit. Together with a degree-weighted selection rule, this bounds the work of each level by its number of edges and stored ball entries. More generally, a ball need not be hit whenever the analysis only uses clusters up to a radius bounded by a certificate that the algorithm has already found.

Incentives can align private interests with collective goals, but their design becomes intertwined with learning when agents' preferences are unknown. This paper develops an adaptive incentive-design framework for multi-agent linear-quadratic dynamic games, where a system planner can simultaneously learn agents' private preferences and steer their strategic behavior towards the social optimum. To this end, we first characterize the complete family of affine incentives that implement the social optimum as the unique open-loop Nash equilibrium while ensuring exact budget balance. When agents' preferences are unknown, we develop an online algorithm to estimate their preference parameters and establish strong consistency of the estimates under a weak trajectory-dependent excitation condition, without requiring persistent excitation or i.i.d. trajectories. We then integrate this estimator into an active-learning incentive mechanism that combines certainty-equivalent coordination with decaying random probing. We prove that the probing generates sufficient information through the agents' equilibrium responses and the system dynamics, ensuring consistent preference learning from the resulting closed-loop trajectories. We show that the induced trajectories converge almost surely to their socially optimal counterparts, while cumulative social-cost regret relative to the full-information oracle and cumulative net-budget imbalance grow sublinearly. Simulations of a transboundary pollution game demonstrate the effectiveness of the proposed framework.

Speech recognition on low-cost microcontrollers is, in practice, limited to keyword spotting or a fixed list of commands. We present Oído, an inference engine that runs open-vocabulary English speech recognition entirely on an ESP32-S3 (240 MHz dual-core Xtensa LX7, 512 KB of SRAM, 8 MB of PSRAM, 16 MB of flash, no neural accelerator). Oído executes a 13M-parameter Conformer-CTC model with int8 weights, per-frame dynamic int8 activations and int8 attention on the chip's SIMD instructions, with a schedule designed around flash and PSRAM bandwidth. On LibriSpeech test-clean/test-other the on-device arithmetic reaches 3.70%/8.23% word error rate (WER), within 0.12 points of full precision, and 3.33%/7.15% with an on-chip 1.3M-parameter language model; a transducer head gives 2.99%/6.67% and an int4 variant fits in 8.3 MB. This is 2.3-2.8x fewer errors than the vendor's recognizer on the same chip, and fewer than Whisper tiny.en and Moonshine tiny in full precision on a laptop, including in real noise and reverberation. Measured on a physical board, the firmware runs at 1.97x the duration of the audio on one core, with transcripts identical to an instruction-exact emulator: accurate, but not yet real time (a 4 s command is transcribed about 8.7 s after it ends). The two-core schedule gives wrong results on silicon and is disabled. A streaming variant shows partial text while the speaker talks and delivers the final text 4.1-11.1 s after clips of 3.6-10.6 s (7.2-22.4 s in utterance mode) at 4.86%/11.03% WER; a Spanish model makes 34-60% fewer errors than multilingual Whisper tiny. Code and models are released.

Indoor climbing creates access barriers for blind and low-vision (BLV) climbers because route understanding and in-climb next-hold localization are typically visually mediated and socially dependent. We present HoldLight, a workflow-centered accessibility system combining layered pre-climb route preview, concise real-time next-hold cueing, and explicit fallback under uncertainty. We grounded the design through formative inquiry with BLV climbers, refined the interaction via pilot testing, and conducted a controlled within-subjects evaluation comparing HoldLight with a controlled on-demand human-support condition derived from current support practices. Findings provide initial evidence that structured support can improve pre-climb route understanding, support recovery from next-hold uncertainty, and reduce immediate human re-entry for routine route-information clarification under controlled conditions. HoldLight contributes a bounded accessibility workflow, demonstrating where structured route-information is helpful, where explicit fallback is necessary, and how technical assistance can integrate with retained human safety oversight.

Compression schemes provide a general framework for deriving data-dependent risk bounds for learning algorithms. For learning rules based on preferent compression, existing results upper-bound the risk in terms of a sample-dependent complexity $\boldsymbol{k}$, defined as the cardinality of the compression augmented by the training examples on which the reconstructed hypothesis is inappropriate. These results, however, yield a margin of order $O(\sqrt{\ln N/N})$ above $\boldsymbol{k}/N$. In this paper, we establish that the logarithmic factor is not intrinsic. We prove that, for every fixed confidence level $1-\delta$, the risk is bounded with probability at least $1-\delta$ by $$ \frac{\boldsymbol{k}}{N} + O\left(\frac{1}{\sqrt{N}}\right), $$ uniformly over all possible values of $\boldsymbol{k}$. Thus, risk estimates for learning algorithms based on preferent compression attain the same law-of-large-numbers rate as the empirical estimate of the probability of a fixed event. We also show that the leading term $\boldsymbol{k}/N$ is necessary and investigate its relation to the empirical risk. In particular, we identify regimes in which the empirical risk retains the $O(1/\sqrt{N})$ convergence rate, as well as regimes in which it converges more slowly or ceases to be consistent, while the complexity-based estimator preserves the $O(1/\sqrt{N})$ rate.

Recent accent classification systems can achieve high accuracy in standard in-corpus evaluations. However, when speakers overlap between the training and test sets, such results may overestimate model robustness compared with real-world deployment scenarios. This study revisits the robustness of DistilHuBERT-based accent classification by using L2-ARCTIC and multiple external English accent datasets. The comparison covers file-level and speaker-disjoint splits, along with balanced external-data training and teacher-student pseudo-labeling. The large performance gap between in-corpus and external evaluations suggests that file-level evaluation substantially overestimates performance, while speaker-disjoint evaluation can provide a more realistic estimate of the generalization to unseen speakers. In this study, models trained on external data alone do not reliably generalize to L2-ARCTIC. Among the tested configurations, balanced external-data training offers the most promising supervised setting. The pseudo-labeling experiments further indicate that confidence-based filtering alone is insufficient under domain mismatch. Overall, this study suggests that robust accent classification evaluation should simultaneously include speaker-disjoint evaluation and cross-corpus testing.

Collective Autonomous Vehicle (AV) route choices in real-world traffic networks can create externalities for Human-Driven Vehicles (HDVs), as routes that benefit AVs may increase travel times for other road users. To address this, we introduce a social component, the marginal cost, to the selfish travel time AV reward. Marginal cost captures an AV's impact on other road users by comparing their travel times with those in a counterfactual simulation in which the AV is removed. We study AV route choice in mixed-autonomy traffic settings using Multi-Agent Reinforcement Learning (MARL), where AVs learn routing policies through repeated interactions with the traffic environment. Experiments on the real-world networks of Ingolstadt and Provins show that incorporating marginal cost into the AV reward not only limits the increase in HDV travel times and shortens AV travel times relative to selfish AV routing, but also improves overall system performance. In Ingolstadt, selfish routing of 40% AVs increases the travel times of other vehicles by 1.7%, and this drops to 1.1% when the social reward is included. Interestingly, not only does overall system performance improve, but the AVs are also better off when acting socially.This effect is consistent across tested cities, penetration rates, and replications, yet when the weight of the social component in the reward increases, the system fails to reach the desired performance. For a more scalable approach to social AV routing in mixed-autonomy traffic, we compute the marginal cost only for the AVs whose route choices most affect the traffic system. This reduces the number of counterfactual simulations and achieves similar performance at a significantly lower computational cost. Our results suggest that AV routing should go beyond selfish rewards and, for the benefit of all traffic participants, include a social component in its objective.

Reinforcement learning (RL) has emerged as a key paradigm for improving large language model (LLM) agents, enabling stronger reasoning, planning, and tool-use in interactive environments. However, existing agent RL methods remain fundamentally constrained by an on-policy reasoning bottleneck: when tasks exceed the model's current capability boundary, effective trajectories become extremely sparse and training often collapses. We introduce Rollout-Time Intervention (RTI), a technique designed to elicit successful trajectories beyond the policy's intrinsic exploration capacity. RTI operates by leveraging reward-side evidence (i.e., evaluation rubrics) during rollout to convert erroneous trajectories into successful ones, providing effective learning signals for tasks that the policy rarely solves on its own. Although intervention trajectories are inherently off-policy, we find that they can effectively guide policy optimization when trajectories with excessive distributional shifts are filtered out. Extensive experiments across five benchmarks demonstrate that RTI consistently outperforms RL baselines, enabling the post-trained policy to approach the performance of substantially stronger contemporary LLMs and, more importantly, transcend its original capability boundary. The code is available at https://github.com/Alibaba-NLP/qqr.

Anomaly detection, the identification of collision events that deviate from Standard Model expectations, is one of the most promising model-agnostic strategies for new-physics searches at the Large Hadron Collider. Searching for rare events in learned low-dimensional representations removes the need to commit to hand-crafted observables a priori, but sacrifices the immediate physical interpretation those observables provide. For anomaly detection to enter a discovery pipeline, flagged events must therefore be diagnosed locally, both to distinguish potential new physics from mismodeling or detector effects and to identify observables for targeted follow-up searches. We evaluate supervised and self-supervised neural embedding strategies on COLLIDE-2V, a 750-million-event dataset simulated under realistic High-Luminosity LHC conditions, using the reconstruction error of an autoencoder trained in the learned latent space as the anomaly score. Across the configurations and benchmark signals considered, supervised objectives, which exploit simulation process labels during training, generally outperform their self-supervised counterparts and, in their best configurations, improve on an autoencoder trained directly on kinematic features, raising the signal efficiency at 10% background acceptance on all held-out processes. We further introduce a local diagnostic pipeline that fits a Gaussian Mixture Model to the SM latent space, profiles each component through high-level observables without process labels, localizes the flagged events, and ranks observables by their local discriminating power. The ranking recovers the defining kinematics of the held-out processes, while correlating the anomaly score with the same observables quantifies how closely the score tracks them, revealing cases where the most discriminating observable is not the one the score follows most closely.

Photoplethysmography (PPG) is increasingly recorded by wearable and clinical devices at enormous scale, and a growing number of PPG foundation models (FMs) have been developed to compute reusable representations for various downstream tasks. These models are difficult to compare: different evaluations use different datasets, task definitions, preprocessing, downstream heads, hyperparameter tuning, and baselines. We present PPG-Bench, a unifying open-source benchmark evaluating five public PPG FMs across nine open-access datasets and 26 tasks spanning cardiovascular state estimation, activity, sleep, screening, and identification. We evaluate models under four downstream regimes of increasing adaptability (linear probing, random forests, frozen MLP probes, and end-to-end finetuning), with model-specific hyperparameter optimization and subject-disjoint cross-validation. We also evaluate against three controls: random/constant baseline, a deterministic pulse-rate-variability (PRV) feature set, and a generic time-series FM (MOMENT). Across all 68 task-regime combinations, 35.3% show no meaningful FM advantage over the random/constant baseline, and in 83.8% no PPG-specific FM outperforms both PRV and MOMENT. We also find that added downstream head capacity generally improves performance, but can cause significant degradation if the model is performing at chance/constant level. Current PPG FMs therefore cannot yet claim to provide generalised and transferrable representations of PPG signals, and future evaluation in this field should include checks against simple, non-learned controls. Code for the benchmark and instructions to evaluate more models can be found here - https://github.com/mannas-ai/ppg-bench.

This paper addresses the grid quantization error of deep-learning (DL)-based direction-of-arrival (DOA) estimation. Classification- based DOA methods discretize the continuous angular space into a grid, so that with a peak-based read-out, the accuracy and resolution are limited by the grid step and by the width of the label peak. To address this issue, we propose a sigmoid label that encodes the sources' DOAs as the sum of angle-shifted monotonic sigmoid functions. Differentiating the output of a CNN-Transformer network trained with the sigmoid label produces sharp spectral peaks at the source DOAs. Combined with a moderately refined grid step of 0.25 degrees, the CNN-Transformer with the proposed sigmoid label attains sub-degree accuracy and a 1-degree resolution, breaking the limit of the grid quantization error. Under the unified CNN-Transformer backbone, we compare the proposed Sigmoid label with the Gaussian label, the 0-1 hard label, and the binary cross-entropy (BCE) baseline. Numerical results demonstrate that the Sigmoid label attains the highest accuracy and superior resolution compared to other labels.