arXiv:2602.10234v3 Announce Type: replace Abstract: Stop-and-go traffic waves, a major form of freeway congestion, impose severe and persistent adverse impacts, including reduced traffic efficiency, increased safety risks, and elevated vehicle emissions. Among various freeway traffic management strategies, jam-absorption driving (JAD), in which a dedicated vehicle performs "slow-in" and "fast-out" maneuvers before being captured by a stop-and-go wave, has been proposed as a promising approach to suppressing the propagation of such waves. However, most existing JAD strategies remain impractical, primarily due to the lack of consideration of implementation vehicles and operational conditions. Inspired by real-world observations of police-car swerving behavior, this paper first introduces the Single-Vehicle Double-Detector Jam-Absorption Driving (SD-JAD) problem and then proposes a practical JAD strategy based on a definition of the JAD Triangle, transforming such behavior into a traffic control strategy capable of suppressing the propagation of an isolated stop-and-go wave. Five key parameters that significantly affect the proposed strategy, namely JAD speed, inflow traffic speed, wave width, wave speed, and in-wave speed, are identified and systematically analyzed. Using a SUMO-based simulation as an illustrative example, we further demonstrate how these parameters can be measured in practice using only two stationary roadside traffic detectors. The results show that the proposed JAD strategy successfully suppresses the propagation of a stop-and-go wave without triggering secondary waves. This paper is expected to take a significant step toward the practical implementation of JAD, advancing it from a theoretical concept to a feasible and deployable traffic management strategy.
Science Journals
arXiv:2602.18020v2 Announce Type: replace Abstract: Vision-Language-Action (VLA) models leverage pretrained Vision-Language Models (VLMs) as backbones to map images and instructions to actions, demonstrating remarkable potential for generalizable robotic manipulation. To enhance performance, existing methods often incorporate extra observation cues (e.g., depth maps, point clouds) or auxiliary modules (e.g., object detectors, encoders) to enable more precise and reliable task execution, yet these typically require costly data collection and additional training. Inspired by the finding that Feed-Forward Network (FFN) in language models can act as "key-value memory", we propose Uncertainty-aware Observation Reinjection (UAOR), an effective, training-free and plug-and-play module for VLA models. Specifically, when the current language model layer exhibits high uncertainty, measured by Action Entropy, it reinjects key observation information into the next layer's Feed-Forward Network (FFN) through attention retrieval. This mechanism directly augments the hidden states with observation evidence at high-uncertainty layers, enabling more accurate and reliable action generation. Comprehensive experiments show that our method consistently improves diverse VLA models across simulation and real-world tasks with minimal overhead. Notably, UAOR eliminates the need for additional observation cues or modules, making it a versatile and practical plug-in for existing VLA pipelines. The project page is at https://uaor.jiabingyang.cn.
arXiv:2602.18620v2 Announce Type: replace Abstract: Task-oriented Vehicle-to-Everything (V2X) networks have recently been proposed to scalably support the large-scale deployment of connected vehicles. In task-oriented V2X networks, vehicles select the content of the transmitted messages based on its relevance to the intended receivers. However, estimating relevance can be challenging, especially in highly dynamic and complex driving scenarios. Relevance estimation errors may cause a transmitting vehicle to share incomplete information, omitting relevant data that is critical for the intended receivers' situational awareness. This work numerically demonstrates that task-oriented V2X networks exhibit an inherent resilience to incomplete information sharing. We show that such resilience guarantees a consistent delivery of relevant information even under high relevance estimation error probability conditions. Furthermore, we show that the fundamental conditions underpinning such inherent resilience can also be encountered outside of the V2X domain - in particular, in other task-oriented networks where multiple transmitters select the content of their messages based on the task-related requirements of a common set of intended receivers.
arXiv:2602.21949v2 Announce Type: replace Abstract: In this paper, we investigate a problem of minimizing total energy consumption for secure federated learning (FL) over wireless edge networks. To address the high computational cost and privacy challenges in conventional FL with neural networks (NN) for resource-constrained users, we propose a novel FL with hyperdimensional computing and differential privacy (FL-HDC-DP) framework. In the considered model, each edge user employs hyperdimensional computing (HDC) for local training, which replaces complex neural updates with simple hypervector operations, and applies differential privacy (DP) noise to protect transmitted model information. We optimize the total energy of computation and communication under both latency and privacy constraints. We formulate the problem as an optimization that minimizes the total energy of all users by jointly allocating HDC dimension, transmission time, system bandwidth, transmit power, and CPU frequency. To solve this problem, a sigmoid-variant function is proposed to characterize the relationship between the HDC dimension and the convergence rounds required to reach a target accuracy. Based on this model, we develop two alternating optimization algorithms, where closed-form expressions for time, frequency, bandwidth, and power allocations are derived at each iteration. Since the iterative algorithm requires a feasible initialization, we construct a feasibility problem and obtain feasible initial resource parameters by solving a per round transmission time minimization problem. Simulation results demonstrate that the proposed FL-HDC-DP framework achieves up to 83.3% total energy reduction compared with the baseline, while attaining about 90% accuracy in approximately 3.5X fewer communication rounds than the NN baseline.
arXiv:2602.22919v2 Announce Type: replace Abstract: Cardiac cine magnetic resonance imaging (MRI) is central to functional cardiac assessment, yet a full current cine sequence may not always be directly available at the point of analysis. We introduce Chain of Flow (COF), an electrocardiography (ECG)-conditioned framework that combines patient-specific MRI and current ECG for subject-specific 4D cardiac cine generation. On the UK Biobank dataset, COF achieves strong image-level fidelity and downstream function-oriented performance on a shared same-visit evaluable benchmark. Multi-slice and multi-resolution analyses indicate stable structural generation quality across the short-axis stack and heterogeneous acquisition resolutions. Controlled phase-robustness analyses across resampled input MRI phases further provide same-visit proxy support for patient-specific MRI plus current ECG when a target MRI phase is not directly observed. A cross-visit route provides exploratory serial evidence, with the clearest gains in current-facing region-of-interest readout. Disease-category functional audits, case-level volume-trajectory evidence review further delineate where the current patient-specific MRI plus ECG formulation remains stable for anatomy-aware downstream cardiac analysis. Code is available at https://anonymous.4open.science/r/COF-paper-release-C88B.
arXiv:2603.03292v3 Announce Type: replace Abstract: Large Language Models (LLMs) exhibit high reasoning capacity in medical question-answering, but their tendency to produce hallucinations and outdated knowledge poses critical risks in healthcare fields. While Retrieval-Augmented Generation (RAG) mitigates these issues, existing methods rely on noisy token-level signals and lack the multi-round refinement required for complex reasoning. In this paper, we propose MA-RAG (Multi-Round Agentic RAG), a framework that facilitates test-time scaling for complex medical reasoning by iteratively evolving both external evidence and internal reasoning history within an agentic refinement loop. At each round, the agent transforms semantic conflict among candidate responses into actionable queries to retrieve external evidence, while optimizing history reasoning traces to mitigate long-context degradation. MA-RAG extends the self-consistency principle by leveraging the lack of consistency as a proactive signal for multi-round agentic reasoning and retrieval, and mirrors a boosting mechanism that iteratively minimizes the residual error toward a stable, high-fidelity medical consensus. Extensive evaluations across 7 medical Q&A benchmarks show that MA-RAG consistently surpasses competitive inference-time scaling and RAG baselines, delivering substantial +6.8 points on average accuracy over the backbone model. Our code is available at https://github.com/NJU-RL/MA-RAG.
arXiv:2603.10416v2 Announce Type: replace Abstract: Recently, Ryu et al. showed that two broadened bands connected by a set of four Einstein-coefficient spectra for stimulated and spontaneous single-photon transitions will obey detailed balance at equilibrium if the spectra satisfy generalized Einstein relations. Here, quantum mechanical expressions for Einstein-coefficient spectra are obtained in the electric-dipole approximation using an intramolecular Boltzmann distribution and the quantized field operators in isotropic, dispersive media of Nienhuis and Alkemade. These expressions suggest relationships between Einstein-coefficient spectra and dipole-strength spectra. The electrodynamic relationship between the spectral density for electromagnetic energy and the spectral density for the square of the electric field is developed and used to define dipole-strength spectra in terms of conditional transition probabilities per unit time. These rigorously relate dipole-strength spectra to Einstein-coefficient spectra, thus establishing quantum formulas for dipole-strength spectra and new generalized Einstein relations between dipole-strength spectra. For transitions between two bands, the dipole-strength spectra depend on a single total dipole strength but replace Einstein's degeneracy ratio and transition frequency with a change in standard chemical potential and a single underlying lineshape that is manifested differently in the four spectra. At equilibrium, the relations specify the Stokes' shift between forward and reverse transitions. The relationships between dipole-strength spectra, spontaneous emission spectral densities and stimulated transition cross sections depend on the refractive index, the dielectric constant, and the local field, but not on the derivative of the refractive index. The broadband relationships reduce to known relationships for narrow spectra inside materials and for line spectra in vacuum.
arXiv:2603.12666v2 Announce Type: replace Abstract: Retrosynthesis prediction aims to identify reactants that can synthesize a given product molecule. Although molecular large language models (LLMs) have recently shown promising results, most existing methods either generate reactants directly or provide only generic product-level analysis, without explicitly reasoning about bond-disconnection strategies that justify specific reactant choices. This paper proposes RetroReasoner, a retrosynthetic reasoning model that captures chemists' strategic disconnection-based thinking. RetroReasoner is trained with supervised fine-tuning and reinforcement learning. For supervised fine-tuning, SyntheticRetro generates structured disconnection rationales paired with reactant predictions. For reinforcement learning, a round-trip reward evaluates predicted reactants by passing them through a forward synthesis model and rewarding predictions that reconstruct the original product. RetroReasoner can also be applied to multi-step retrosynthetic planning by incorporating it into a parallelized Monte Carlo tree search framework, reducing search time while increasing the number and diversity of valid synthetic pathways. Experimental results show that RetroReasoner outperforms prior baselines, including not only molecular LLMs but also retrosynthesis-specific expert models, and generates a broader range of feasible reactant proposals, especially for challenging reaction instances.
arXiv:2603.13431v3 Announce Type: replace Abstract: Computational antibody design has seen rapid methodological progress, with dozens of deep generative methods proposed in the past three years, yet the field lacks a standardized benchmark for fair comparison and model development. These methods are evaluated on different SAbDab snapshots, non-overlapping test sets, and incompatible metrics, and the literature fragments the design problem into numerous sub-tasks with no common definition. We introduce CHIMERA-Bench: (CDR Modeling with Epitope-guided Redesign), a unified benchmark built around a single canonical task: epitope-conditioned CDR sequence-structure co-design. CHIMERA-Bench provides three components. The first is a curated, deduplicated dataset of 2,922 antibody-antigen complexes with epitope and paratope annotations. The second is a set of three biologically motivated splits that test generalization to unseen epitopes, unseen antigen folds, and prospective temporal targets. The third is a comprehensive evaluation protocol with five metric groups, including novel epitope-specificity measures. We benchmark eleven methods spanning six generative paradigms and report results across all splits. CHIMERA-Bench is the largest dataset of its kind for the antibody design problem, allowing the community to develop and test novel methods and evaluate their generalizability.
arXiv:2603.15249v2 Announce Type: replace Abstract: This paper establishes tractable bounds of joint source-channel coding with hierarchical sources in the finite blocklength regime. In this setting, both the indirect source and observable source must be reconstructed under correlated distortion constraints, leading to a joint excess-distortion event. First, to build computable tight bounds, we introduce a novel $\mathsf{d}(\cdot)$-functional distortion relaxation, which enables tractable and tight bounding of the joint excess-distortion probability induced by correlated sources. By this approach, the nonasymptotic converse and achievability bounds are given. Second, Gaussian approximations for the proposed bounds are obtained, which are optimal for the transmission of a Gaussian memoryless source over an additive white Gaussian noise channel with mean-square error distortion. The optimal scheme is obtained via a structured analysis that captures the intrinsic tradeoff between semantic and observable reconstructions. Furthermore, for the transmission of Gaussian memoryless sources over AWGN channels, we obtain explicit and computable bounds, by providing a new geometric structure involving three correlated spherical regions. This results extend the classical two-spherical region analysis for a single distortion constraint. Numerical simulations demonstrate that the proposed achievability and converse bounds tightly sandwich the Gaussian approximation and align closely with Monte Carlo numerical results.
arXiv:2603.16020v2 Announce Type: replace Abstract: Adaptive agents operating under uncertainty must do more than optimize task outputs: they must maintain a workable internal state under noise, perturbation, and changing conditions. This paper introduces IRAM-Omega-Q, a computational framework for modeling uncertainty regulation in adaptive agents under stochastic disturbance. The framework combines a quantum-like state representation with closed-loop adaptive control over an internal entropy signal. The quantum-like formalism is used instrumentally: the evolving state is a normalized complex amplitude vector, coherent evolution is propagated exactly as psi(t + Delta t) = exp(-i H Delta t) psi(t), and a derived density matrix supports entropy and coherence-gap analysis. Two causal control orderings are compared. In regulation-first (RF) ordering, adaptive regulation is available before current-cycle disturbance and attenuates incoming exposure; in disturbance-first (DF) ordering, current-cycle disturbance is received before a new regulatory response can be computed, and stabilization acts reactively. Publication-mode, matched-seed simulations show broadly comparable coherence-gap trajectories but lower sustained adaptive gain under RF. Susceptibility maps based on post-burn-in temporal fluctuations further show that DF shifts the critical initial-gain ridge toward larger values across multiple disturbance intervals. These results identify ordering as an architectural determinant of regulatory demand and threshold location within an otherwise shared regime structure.
arXiv:2603.22473v2 Announce Type: replace Abstract: Hybrid language models combine softmax attention with linear-time sequence mechanisms such as state-space or linear-attention layers, but the functional contribution of each component type remains insufficiently characterized. We study component-level ablation in two sub-1B hybrid language models, Qwen3.5-0.8B and Falcon-H1-0.5B, using likelihood-based evaluation, downstream benchmarks, layer-wise interventions, random controls, and representation-level diagnostics. Across the tested models, removing either attention or the alternative sequence-processing pathway substantially degrades performance, indicating that both component types contribute to model behavior. Likelihood metrics are especially sensitive to the linear-attention or state-space pathway, while downstream benchmark degradation depends on task and architecture. Layer-wise ablations show that component importance is position-dependent, with the strongest effects concentrated in early or mid-network components rather than uniformly across depth. Random-removal controls further show that hybrid architectures and same-family Transformer baselines degrade differently under structural perturbation. These results suggest that component ablation is a useful diagnostic for understanding hybrid language model architectures. The findings provide evidence relevant to efficient model design, compression, robustness analysis, and deployment decisions in architectures that combine attention with alternative sequence-processing mechanisms.
arXiv:2603.22793v2 Announce Type: replace Abstract: Classroom AI systems increasingly infer high-level educational states such as engagement, confusion, collaboration, participation, and instructional quality from multimodal and linguistic signals. In multicultural and multilingual classrooms, such inferences can translate culturally situated behavior into stereotyped claims: silence may be read as disengagement, gaze aversion as inattention, code-switching as low proficiency, or indirect help-seeking as confusion. We argue that stereotype-aware classroom AI should separate observable evidence from culturally loaded interpretation and should treat unsupported construct-level claims as safety risks. We introduce NSCR, a culturally grounded neuro-symbolic framework that converts video, audio, ASR, lesson artifacts, and contextual metadata into typed facts with uncertainty, provenance, and cultural scope, then composes them through executable reasoning and policy constraints. We define a taxonomy of stereotype-prone classroom inferences and propose a benchmark agenda covering culture-conditioned state inference, evidence-grounded claim verification, multilingual and code-switched reasoning, collaboration analysis, counterfactual cultural robustness, and culture-conditioned red-teaming. We further specify metrics for stereotype leakage, unsupported attribution, cultural calibration gaps, abstention under cultural ambiguity, and evidence faithfulness. The contribution is methodological: a concrete framework and evaluation agenda for mitigating stereotyped reasoning in classroom AI, with education as a high-stakes, culturally variable deployment setting.
arXiv:2604.04251v2 Announce Type: replace Abstract: Intelligent tutoring systems increasingly rely on reinforcement learning to personalise instruction, yet optimising for observable engagement signals can systematically decouple learner activity from genuine knowledge acquisition. Analysing over 21 million student interactions across two deployed platforms, we find engagement events without corresponding mastery gains occur in 26.5% of interactions on Junyi Academy (72,758 students) and 3.1% on XES3G5M (14,453 students, NeurIPS 2023), confirming this pattern is directly observable in deployed educational technology at scale. We introduce Mastery-Conditioned Constrained Policy Optimisation (MC-CPO), a reinforcement learning framework that addresses this problem structurally. MC-CPO conditions the admissible instructional action space on learner mastery state: a concept becomes available only when prerequisite knowledge meets a mastery threshold, yielding an action space that expands naturally as learners acquire knowledge. Pedagogical safety constraints are enforced by construction, with formal guarantees of structural prerequisite safety, primal-dual convergence, and strict dominance over post-hoc filtering. MC-CPO is the only method to reduce reward hacking severity across all conditions. Mean per-episode mastery gain increases by 18.3% on Junyi Academy and 54.0% on XES3G5M relative to all baselines, while competitive engagement performance is maintained. These results support structural constraint modelling as a principled foundation for safer adaptive instructional policies in deployed tutoring systems.
arXiv:2604.10271v4 Announce Type: replace Abstract: In what way could a data breach involving government-issued IDs such as passports, driver's licenses, etc., rival a random voluntary disclosure on a nondescript social-media platform? At first glance, the former appears more significant, and that is a valid assessment. The disclosed data could contain an individual's date of birth and address; for all intents and purposes, a leak of that data would be disastrous. Given the threat, the latter scenario involving an innocuous online post seems comparatively harmless--or does it? From that post and others like it, a forensic linguist could stylometrically uncover equivalent pieces of information, estimating an age range for the author (adolescent or adult) and narrowing down their geographical location (specific country). While not an exact science--the determinations are statistical--stylometry can reveal comparable, though noticeably diluted, information about an individual. To prevent an ID from being breached, simply sharing it as little as possible suffices. Preventing the leakage of personal information from written text requires a more complex solution: adversarial stylometry. In this paper, we explore how performing homoglyph substitution--the replacement of characters with visually similar alternatives (e.g., "h" $\texttt{[U+0068]}$ $\rightarrow$ "h" $\texttt{[U+04BB]}$)--on text can degrade stylometric systems.
arXiv:2604.18050v2 Announce Type: replace Abstract: AlphaGeometry represents a milestone in neuro-symbolic reasoning, yet its architecture faces a log-linear scaling bottleneck within its symbolic deduction engine that limits its efficiency as problem complexity increases. Recent technical reports suggest that current domain-specific languages may be isomorphic as input representations to natural language, interchanging them acts as a performance-invariant transformation, implying that current neural guidance relies on superficial encodings rather than structural understanding. This paper addresses this representation bottleneck by proposing a logic-to-topology encoding designed to reveal the structural invariants of a model's latent space under a transformation of its input space. By leveraging the Logic of Observation, we utilize the duality between provability in observable theories and topologies to propose a logic-to-topology encoder for the input space. We introduce the concept of the "topological dual of a dataset", a transformation that bridges formal logic, topology, and neural processing. This framework serves as a Rosetta Stone for neuro-symbolic AI, providing a principled pathway for the mechanistic interpretability of how models navigate complex discovery paths.
arXiv:2604.18347v2 Announce Type: replace Abstract: Vision Language Models (VLMs) achieved rapid progress in the recent years. However, despite their growth, VLMs development is heavily grounded on English, leading to two main limitations: (i) the lack of multilingual and multimodal datasets for training, and (ii) the scarcity of comprehensive evaluation benchmarks across languages. In this work, we address these gaps by introducing a new comprehensive suite of resources for VLMs training and evaluation spanning five European languages (English, French, German, Italian, and Spanish). We adopt a regeneration-translation paradigm that produces high-quality cross-lingual resources by combining curated synthetic generation and manual annotation. Specifically, we build Multi-PixMo, a training corpus obtained regenerating examples from Pixmo pre-existing datasets with permissively licensed models: PixMo-Cap, PixMo-AskModelAnything, and CoSyn-400k. On the evaluation side, we construct a set of multilingual benchmarks derived translating widely used English datasets (MMbench, ScienceQA, MME, POPE, AI2D). We assess the quality of these resources through qualitative and quantitative human analyses, measuring inter-annotator agreement. Additionally, we perform ablation studies to demonstrate the impact of multilingual data, with respect to English only, in VLMs training. Experiments, comprising 3 different models show that using multilingual, multimodal examples for training VLMs aids is consistently beneficial on non-English benchmarks, with positive transfer to English as well.
arXiv:2604.19845v4 Announce Type: replace Abstract: Self-modification is routinely treated as constitutive of artificial superintelligence (\textbf{SI}), yet modification is a relative action requiring a \emph{supplement} outside the operation. We formalise this on an associative operator algebra $\mathcal{A}$ with update operator $\hat U$, difference operator $\hat D$, and self-representation operator $\hat R$, identifying the supplement with $\operatorname{Comm}(\hat U)$. A propagation theorem shows $[\hat U,\hat R]$ decomposes through $[\hat U,\hat D]$, so non-commutation propagates to self-representation. The liar paradox is the rank-one case $[\hat T,\Pi_L]=0$, and \emph{class $\mathbf{A}$} systems, in which $\hat U$ acts on $\hat D$, reproduce it at system scale, yielding a structure coinciding with Priest's inclosure schema and Derrida's \emph{diff\'erance}. Our results show that the strong self-modification taken to define superintelligence may undermine the persistent identity upon which such systems are premised.
arXiv:2604.22482v2 Announce Type: replace Abstract: While feed-forward 3D reconstruction models have advanced rapidly, they still exhibit degraded performance on panoramas due to spherical distortions. Moreover, existing panoramic 3D datasets are predominantly collected with 360 cameras fixed at discrete locations, resulting in discontinuous trajectories. These limitations critically hinder the development of panoramic feed-forward 3D reconstruction, especially for the multi-view setting. In this paper, we present Holo360D, a comprehensive dataset containing 109,495 panoramas paired with registered point clouds, meshes, and aligned camera poses. To our knowledge, Holo360D is the first large-scale dataset that provides continuous panoramic sequences with accurately aligned high-completeness depth maps. The raw data are initially collected using a 3D laser scanner coupled with a 360 camera. Subsequently, the raw data are processed with both online and offline SLAM systems. Furthermore, to enhance the 3D data quality, a post-processing pipeline tailored for the 360 dataset is proposed, including geometry denoising, mesh hole filling, and region-specific remeshing. Finally, we establish a new benchmark by fine-tuning 3D reconstruction models on Holo360D, providing key insights into effective fine-tuning strategies. Our results demonstrate that Holo360D delivers superior training signals and provides a comprehensive benchmark for advancing panoramic 3D reconstruction models. Datasets and Code will be made publicly available.
arXiv:2604.22566v2 Announce Type: replace Abstract: Despite their ubiquity, the rich physics present in a plasma sheath has inhibited the development of a generally applicable description of this critical region. The present study utilizes a physics-informed neural network (PINN) to evaluate a hierarchy of models of the plasma sheath. Unlike traditional deep learning methods, PINNs use the governing PDEs to constrain the predictions of a neural network, and thus do not require any experimental or simulation data to train. In this work, we utilize a PINN to identify the parametric solution to fluid models of different physics fidelity of the plasma sheath. While the offline training time of the PINN is often longer than a traditional solver, once trained, the PINN is able to efficiently predict the sheath profiles across a broad range of parameter regimes, thus yielding an effective surrogate of the plasma sheath.
arXiv:2604.24594v3 Announce Type: replace Abstract: As large language models (LLMs) evolve into agentic problem solvers, they increasingly rely on external, reusable skills to handle tasks beyond their native parametric capabilities. In existing agent systems, the dominant strategy for incorporating skills is to explicitly enumerate available skills within the context window. However, this strategy fails to scale: as skill corpora expand, context budgets are consumed rapidly, and the agent becomes markedly less accurate in identifying the right skill. To this end, this paper formulates Skill Retrieval Augmentation (SRA), a new paradigm in which agents dynamically retrieve, incorporate, and apply relevant skills from large external skill corpora on demand. To make this problem measurable, we construct a large-scale skill corpus and introduce SRA-Bench, the first benchmark for decomposed evaluation of the full SRA pipeline, covering skill retrieval, skill incorporation, and end-task execution. SRA-Bench contains 5,400 capability-intensive test instances and 636 manually constructed gold skills, which are mixed with web-collected distractor skills to form a large-scale corpus of 26,262 skills. Extensive experiments show that retrieval-based skill augmentation can substantially improve agent performance, validating the promise of the paradigm. At the same time, we uncover a fundamental gap in skill incorporation: current LLM agents tend to load skills at similar rates, regardless of whether a gold skill is retrieved or whether the task actually requires external capabilities. This shows that the bottleneck in skill augmentation lies not only in retrieval but also in the base model's ability to determine which skill to load and when external loading is actually needed. These findings position SRA as a distinct research problem and establish a foundation for the scalable augmentation of capabilities in future agent systems.
arXiv:2604.25193v2 Announce Type: replace Abstract: Inverse design has transformed vast physical parameter spaces into a substrate for emergent functionality, raising the tantalizing prospect of relocating intelligence from the digital domain into the physical world itself. Nowhere is this prospect more consequential than in sensing, where the analog-to-digital interface imposes a fundamental bottleneck: information not captured by the hardware is irrevocably lost to any downstream algorithm. Existing approaches improve information capture through either sensor hardware optimization or adaptive measurement strategies operating on fixed hardware, but rarely both in concert. A principled migration of intelligence from digital to physical demands their joint optimization: the sensing geometry must be co-designed with a policy that determines what to measure next. We formulate this co-design as joint dynamic programming (joint-DP), a unified optimization over sensor geometry and a Bellman-optimal adaptive measurement policy. The outer hardware gradient is obtained through differentiable dynamic programming with a sharp Bellman maximum. A hierarchy of relaxations extends the framework from small discrete POMDPs to freeform photonic topologies with more than $10^5$ design pixels.
arXiv:2604.26333v2 Announce Type: replace Abstract: Distributional learning provides a useful framework for studying the learnability of structured languages from positive data. In this paper, we extend this framework to graph languages generated by fixed-interface clause systems (FICSs). We formulate FICSs explicitly and study the corresponding learning problem under positive presentations and membership queries. We consider a bounded class of graph languages satisfying the finite context property (FCP) under a bounded-degree assumption. The bounds are expressed by the degree bound $\Delta$ together with five structural parameters $m,s,t,w$, and $d$, which control the clause-system structure, interface ranks, and local head-frame complexity. The learning algorithm constructs hypotheses from ordered boundary representations induced by the observed positive examples. These representations make explicit the interface information needed to compare contexts and to test candidate clauses by membership queries. We prove that target contexts eventually appear in the observed sample, target clauses are reconstructed over the corresponding predicate representatives, and spurious non-fact clauses are eventually excluded. Consequently, for every fixed parameter tuple, the target language is identifiable in the limit from positive data and membership queries. We also prove that the learner has polynomial-time update on $\FICSLFCP_{\Delta}(m,s,t,w,d)$: at each stage, only polynomially many ordered boundary representations, predicate symbols, clause candidates, and membership queries are needed. Overall, the paper gives a parameterized reformulation of distributional learning for interface-based graph languages in a fixed-interface setting.
arXiv:2604.26985v2 Announce Type: replace Abstract: Masked diffusion models (MDMs) generate discrete sequences by iterative denoising under an absorbing masking process. In standard masked diffusion, if a token remains masked after a reverse update, the model discards its clean-state prediction for that position. Thus, still-masked positions must be repeatedly inferred from the mask token alone. This design choice limits cross-step refinement. To address this limitation, this paper proposes a simple, yet effective, post-training adaptation for MDMs that conditions each denoising step on the model's own previous clean-state predictions. The resulting method, called Self-Conditioned Masked Diffusion Models (SCMDM), requires minimal architectural change, does not introduce a recurrent latent-state pathway, does not rely on an auxiliary reference model, and adds no extra denoiser evaluations during sampling. This is an important departure from partial self-conditioning approaches which requires expensive model training from scratch. In particular, the paper shows that partial self-conditioning, including the commonly used 50% dropout strategy for training self-conditioned models from scratch, is suboptimal in the post-training regime. Instead, once the model's self-generated clean-state estimates become informative, the specialization to refinement is preferable to mixing conditional and unconditional objectives. SCMDM is evaluated across multiple domains, demonstrating consistent improvement over vanilla MDM baselines, achieving nearly a 50% reduction in generative perplexity on OWT-trained models (42.89 to 23.72), alongside strong improvements in discretized image synthesis quality, small molecular generation, and enhanced fidelity in genomic distribution modeling.
arXiv:2604.26993v2 Announce Type: replace Abstract: We study gradient descent for rank-1 matrix factorization through a state-dependent Lyapunov perspective. The central object is a parameterized quadratic certificate $I(\delta;\,\cdot)$ whose boundary-inward property induces a monotone state parameter $\delta_t$, thereby certifying that the trajectory is confined to a shrinking family of level sets. For certified initializations below the critical step size, this mechanism proves convergence to global minimizers. Above the critical step size, the same monotone-state mechanism instead leads to a balanced terminal regime; for a range of post-critical step sizes, the reduced dynamics exhibit period-2 behavior consistent with edge-of-stability phenomena. We further show that the scalar certificate is not an ad hoc algebraic construction: under structural axioms and a natural state-parameter normalization, it is uniquely determined by the monotonicity mechanism. Numerical experiments suggest that this state-dependent Lyapunov mechanism persists beyond the proved cases, including two-dimensional rank-1 approximation and quartic augmentations of scalar factorization.