arXiv:2606.18375v3 Announce Type: replace Abstract: World foundation models (WFMs) are powerful simulators, yet they predominantly operate in a single-view setting and lack the multi-view 3D consistency required for robotic manipulation. While robotic systems rely on multiple cameras (egocentric, eye-to-hand, and wrist-mounted) for policy learning, current multi-view world models simply concatenate view tokens without explicit geometric reasoning. This causes cross-view object drift, depth inconsistency, and texture misalignment. We trace these failures to two deficiencies: the absence of an explicit inter-view communication mechanism and the lack of a 3D geometric prior. We argue that resolving both simultaneously is necessary and sufficient. To address this, we present PAIWorld, a framework that augments diffusion-transformer world models via three core components: (1) Geometry-Aware Cross-View Attention blocks that establish an explicit pathway across views, (2) Geometric Rotary Position Embedding that encodes camera ray directions and extrinsic poses into the attention mechanism, and (3) Latent 3D-REPA, which distills 3D-aware features from frozen 3D foundation models to ensure 3D consistency. Built upon a DiT-based world foundation model, PAIWorld achieves state-of-the-art multi-view 3D consistency on robotic manipulation benchmarks, ranking 1st on the WorldArena leaderboard and 2nd on the AgiBot-Challenge2026 leaderboard, while enabling downstream applications such as model-based planning, world action models, and multi-view policy post-training.
Science Journals
arXiv:2606.22524v2 Announce Type: replace Abstract: We develop a risk-aware information theory by replacing expectation with expectiles, introducing expectile entropy, divergence, and mutual information. These quantities exhibit behaviors impossible under Shannon's risk-neutral framework, including negative divergence under risk-seeking regimes and a fundamental separation from classical mutual information. In multiuser systems, the framework naturally induces a mean-field-type game theory of information exchange, where achievable rate regions become endogenous to heterogeneous risk-sensitivity indices. Our results reveal that Shannon information alone cannot quantify the extreme risks driving advanced machine intelligence, establishing a foundation for risk-aware communication, learning, collective intelligence, and safe autonomous systems.
arXiv:2506.03759v2 Announce Type: replace-cross Abstract: In this paper, we study stabilizability of discrete-time switched linear systems where the switching signal is considered as an arbitrary external input (and not a control variable). We characterize feedback stabilization via a hierarchy of necessary and sufficient linear matrix inequalities (LMIs) conditions based on novel graph structures. We analyze both the cases in which the controller has (or has not) access to the current switching mode, the so-called mode-dependent and mode-independent settings, providing specular results. Moreover, our approach provides explicit piecewise-linear and memory-dependent linear controllers, highlighting the connections with existing stabilization approaches. The effectiveness of the proposed technique is finally illustrated with the help of some numerical examples.
arXiv:2605.17917v2 Announce Type: replace-cross Abstract: We review the theoretical development and modern applications of the Evans-Sackmann hydrodynamic model for lateral transport in supported fluid membranes. We first cover the original formulation, emphasizing the linear momentum decay term that captures membrane-substrate coupling mediated by a thin lubricating fluid layer. This coupling term enables quantitative interpretation of tracer diffusion measurements in supported bilayers. Building on this foundation, we survey theoretical extensions that relax standard boundary conditions at the inclusion perimeter, where inclusions refer to embedded objects such as proteins, lipid domains, or tracer particles within the membrane. We discuss the drag of a disk and a liquid domain, as well as the dynamics of membrane phase separation. We further highlight how the supported-membrane mobility tensor serves as a unifying tool for systematic treatments of correlated diffusion, polymer dynamics, phase separation kinetics, and many-body interactions in quasi-two-dimensional environments. Finally, we discuss recent extensions to active and chiral membranes, where odd viscosity provides a transverse hydrodynamic response and offers a possible route for detecting chirality in two-dimensional fluids.
arXiv:2606.22073v2 Announce Type: replace Abstract: This paper proposes an energy-recycling rimless wheel with spring-clutch legs. The proposed mechanism uses a lockable clutch to store part of the impact-induced elastic energy after foot contact and reinject it in the next gait cycle. First, we develop a hybrid dynamic model of the energy-recycling rimless wheel. Second, numerical simulations are used to examine the dynamics, local stability of periodic gaits, and the Cost of Transport (CoT) of the proposed mechanism. The simulation results show that the proposed mechanism reduces the CoT by up to 16.13% compared with a benchmark viscoelastic-legged rimless wheel with telescopic spring-damper legs. Compared with the rigid rimless wheel, the viscoelastic-legged and energy-recycling models reduce the CoT by more than 50%. The energy-recycling model also maintains locally stable periodic gaits over the tested slope and stiffness ranges. Finally, prototype experiments on an inclined plane are conducted to examine the feasibility of the proposed mechanism. The experimental results show that the proposed rimless wheel achieves passive walking on a shallow 1{\deg} slope, corresponding to a CoT of approximately 0.02. These results suggest that the proposed spring-clutch mechanism can improve the simulated walking efficiency of the energy-recycling rimless wheel, while the prototype experiments support the feasibility of passive walking with the mechanism.
arXiv:2606.22220v2 Announce Type: replace Abstract: Memorization in machine learning models enables high performance on rare in-distribution samples by capturing their atypical patterns. However, it also causes harmful retention of noise and outliers, degrading generalization. While memorization has been extensively studied in both supervised and self-supervised learning in the vision domain, it remains unexplored in multi-modal contrastive learning. We address this gap by introducing MultiMem, the first metric designed to quantify memorization in multi-modal contrastive learning. Through our systematic analysis, we demonstrate that cross-modal semantic misalignment has the strongest influence on memorization, with text being the dominant modality driving memorization, followed by video, image, and audio. We show that targeted augmentations applied across all modalities effectively reduce memorization as measured by our MultiMem metric and improve model performance. Overall, this work establishes the first framework for measuring and mitigating memorization in multi-modal contrastive learning, preventing harmful data retention and contributing to higher-performing models.
arXiv:2606.22413v2 Announce Type: replace Abstract: Vibe coding -- accepting LLM-generated source from natural-language intent with minimal review -- is fast and may be adequate for low-criticality consumer software. But for safety-critical systems governed by DO-178C, IEC 61508, or ISO 26262, it offers no path to certification: large language models (LLMs) provide no formal correctness guarantees, and existing remedies target verification-aware languages (Dafny, Verus, Lean) that are scarce in pretraining data and absent from industrial toolchains. This paper closes the gap. We present Forge (Formal method Oriented Refinement loop for GEnerated code): a closed-loop pipeline that guides vibe coding through formal verification using established Model-Driven Engineering (MDE) infrastructure. Through vibe coding, we generate Java source code; our pipeline then extracts -- via model transformations -- formal artefacts in three different formalisms, each checked by a complementary verifier: deductive verification (Dafny), Communicating Sequential Processes (CSP) refinement via the Failures-Divergences Refinement checker (FDR4), and theorem proving using Z-Machines in Isabelle; every verification failure becomes a structured correction prompt that drives the next code-generation iteration. The LLM is the draft generator, the MDE chain is the discriminator, and the developer never has to read the formal models. Empirically, we find that the pipeline produces standards-relevant verification evidence for LLM-generated Java -- a step toward certification.
arXiv:2606.22419v2 Announce Type: replace Abstract: A recent Nature Medicine study reports that general-purpose frontier LLMs outperform specialized retrieval-augmented clinical tools on medical benchmarks, and that retrieval can hurt strong models. We ask the natural follow-up: does structured knowledge-graph (KG) grounding change this, and when does grounding help at all? We contribute two results. First, a reproduction: the study's headline HealthBench score (~88) is the Consensus variant, not full HealthBench, where frontier models and ideal completions both score ~46-47 under a physician-calibrated grader (agreement 82.5%); we reproduce GPT-5.2 Consensus =90.9 and flag a score-deflating grader bug. Second, a knowledge-boundary result. Using a graph+vector engine (samyama-graph) over the public biomedical KG PrimeKG, neither naive triple retrieval nor an agentic natural-language-to-Cypher loop (82% successful queries) improves MedQA across a weak-to-strong model ladder (all |Delta| <= 3.4). On a synthetic counterfactual KG, and on a hybrid benchmark mixing known and novel facts, the identical pipeline lifts out-of-training accuracy from chance to ~100% (+68 to +79) while adding nothing on known facts (a no-LLM arm answers both). Across three regimes (no-knowledge, graph-aided, hybrid), grounding helps only insofar as the decisive fact lies outside the model's training -- public-KG facts are redundant, private and novel data are where it pays -- matching the study's institutional-data caveat.
arXiv:2606.23090v2 Announce Type: replace Abstract: Cross-embodiment data have become central to training robotic foundation models. To leverage such heterogeneous data, we focus on flow-based object manipulation, where robot flows (robot velocity fields) serve as embodiment-agnostic motion representations. Previous studies do not formulate robot flows as dense velocity fields, but as displacements of sparse keypoints, while such velocity fields better match the continuous-time nature of motions. We propose Flow as Flow, a framework that models robot flows as probability flows based on a flow matching formulation. By naturally modeling such velocity fields within this formulation, our method achieves efficient and high-quality robot flow generation. Across standard benchmarks, our method outperforms representative baseline methods on standard metrics, while achieving approximately 33$\times$ faster generation. Furthermore, through real-world experiments evaluating 9 methods with 260 trials per method across 13 manipulation tasks, we show that our method achieves a higher average success rate than the baseline methods. Our project page is available at https://flow-as-flow-u0n5y.kinsta.page.
arXiv:2606.23130v2 Announce Type: replace Abstract: Recent advances in large language models (LLMs) have enabled vibe coding, an emerging software development paradigm in which users create applications primarily through natural-language interactions with AI agents. Due to its low barrier to entry, vibe coding is rapidly gaining adoption in practice. Unlike conventional AI-assisted programming, where developers remain responsible for implementation and code review, vibe coding delegates a substantial portion of development to AI systems. This shift raises a fundamental question: how (in)secure are applications developed through vibe coding? In this paper, we conduct a systematic study of the security of vibe-coded applications. We collect a large corpus of real-world applications developed using popular AI agents and design a vulnerability analysis framework that combines agent-assisted code auditing with human validation. Using this framework, we examine the prevalence, severity, and root causes of vulnerabilities in the deployed vibe-coded applications. Our study reveals several key findings: (1) vibe-coded applications exhibit recurring vulnerability patterns that differ from those commonly observed in conventional software development workflows, including placeholder logic, unfiltered input, and secret exposure; (2) these vulnerabilities arise from systematic limitations of AI agents throughout the vibe-coding lifecycle, such as memory loss, locally optimized objectives and insufficient security knowledge; and (3) while advances in LLM capabilities and improved prompting strategies can reduce the incidence of vulnerabilities, they do not eliminate the underlying security risks. Overall, our study provides an empirical understanding of the security landscape of vibe-coded applications and lays the groundwork for addressing the security challenges introduced by the growing delegation of software development to AI systems.
arXiv:2606.23151v2 Announce Type: replace Abstract: Advances in optical clocks motivate a redefinition of the second, requiring rigorous evaluations of systematic uncertainties and robust consistency among the clocks. Here, we report the full evaluation of the systematic frequency shifts of an $^{27}\mathrm{Al}^+$ single-ion clock, and the measurement of its absolute frequency and frequency ratio with a $^{87}$Sr optical lattice clock at PTB. The evaluated total systematic fractional frequency uncertainty is $1.6\times10^{-18}$, mainly limited by the accuracy of the relevant atomic coefficients and by background gas collisions. The absolute frequency of the clock has been measured to be $\nu_{\mathrm{Al}^+}=1 121 015 393 207 859.19(24)\,$Hz, obtained by comparison with two primary caesium fountain clocks at PTB. The frequency ratio between the Al and Sr optical clocks has been determined to be $\nu_{\mathrm{Al}^+}/\nu_{^{87}\mathrm{Sr}}=2.611 701 431 781 462 668(36)$, limited by the accuracy of the Sr clock. This ratio differs by $8.6\sigma$ and $1.2\sigma$ from the 2021 and 2025 frequency ratio published by the BACON collaboration, respectively. These results represent an important contribution toward a future redefinition of the second using optical clocks, and underscore the importance of independent measurements of clock-candidate frequency ratios across different institutions.
arXiv:2606.23238v2 Announce Type: replace Abstract: Logical reasoning is essential for reliable AI, yet existing benchmarks are largely first-order-logic-centric, focusing on object-level deduction over fixed predicates. This misses many realistic scenarios where models must reason over rules, predicates, functions, constraints, and decision procedures themselves. We introduce HOLMES (Higher-Order Logic Meets real-world Explainable Symbolic reasoning), the first real-world benchmark for higher-order symbolic reasoning in LLMs, containing 1379 instances. Built on higher-order logic, HOLMES pairs natural-language problems with HOL formalizations, ground-truth answers, verifiable reasoning traces, and fine-grained controllable reasoning factors across law and finance. Experiments show that current LLMs still struggle on HOLMES, with an average accuracy of only 50.64% and the best model reaching 59.54%. Our analyses further reveal that high final-answer accuracy can mask shortcut reasoning in conflict-resolution settings, while performance drops sharply under scope-conditioned and compositional reasoning. These findings identify higher-order symbolic reasoning as a key bottleneck for building reliable and verifiable LLMs. The project code and dataset are publicly available at https://github.com/wuyucheng2002/HOLMES.
arXiv:2606.23338v2 Announce Type: replace Abstract: Recent studies Vorticity packing effects on long time turbulent transport in decaying two dimensional incompressible Navier Stokes fluids, Phys. Fluids 38, 045159 (2026) demonstrated that, at a fixed high Reynolds number (Re), the initial vorticity packing fraction (VPF) governs the coupled Eulerian flow evolution and Lagrangian tracer particle transport in decaying two dimensional incompressible Navier Stokes fluids, revealing a strong Eulerian Lagrangian relationship during the nonequilibrium inverse cascade regime and a direct Eulerian Lagrangian correspondence in the late time coherent vortex quasi equilibrium regime, wherein increasing VPF drives transitions from point vortex to patch vortex equilibria and from subdiffusive to superdiffusive transport. In the present work, we investigate how these Eulerian Lagrangian connections evolve across a broad (VPF, Re) parameter space. The results show that the Eulerian Lagrangian relationship remains largely preserved during the nonequilibrium inverse-cascade regime, where transport increases systematically with VPF and remains primarily controlled by VPF despite secondary Re dependent oscillatory modulation. In contrast, the late-time coherent-vortex quasi-equilibrium regime exhibits Eulerian statistical equilibria that remain largely insensitive to Re, while the corresponding tracer-particle transport displays a strong Re dependence characterized by strong oscillatory and nonmonotonic variations across the (VPF, Re) parameter space, substantially weakening the VPF-ordered transport hierarchy observed in the inverse-cascade regime. Consequently, for the parameter range, spatial resolutions, and integration times explored in the present study, the direct Eulerian Lagrangian correspondence identified at fixed high Re is not universally maintained across the broader (VPF, Re) parameter space.
arXiv:2606.23380v2 Announce Type: replace Abstract: AI safety research has focused predominantly on epistemic and physical harms (e.g., misinformation, bias, system reliability) while the risks that arise from AI systems' engagement with human emotional life have remained fragmented and undertheorised. We propose affective safety as a unified class of AI safety concerns grounded in the fact that humans are affective beings. We develop a taxonomy of affective harms and identify recurring harm types: (1) affective self-alienation, (2) fairness and bias harms, and (3) relational harms. We show that their recurrence across system types reflects structural properties of how AI systems engage with human emotion and survey the current safety landscape and show that existing frameworks address affective safety either narrowly or not at all. We conclude by identifying the technical and regulatory challenges specific to this class of harms and argue that affective safety requires dedicated frameworks that engage with cumulative, relational, and identity-level effects.
arXiv:2304.10964v4 Announce Type: replace-cross Abstract: We discuss the Lugiato-Lefever equation and its variant with third-order dispersion, which are mathematical models used to describe how a light beam forms patterns within an optical cavity. It is mathematically demonstrated that the solutions of these equations follow the Talbot effect, which is a phenomenon of periodic self-imaging of an object under certain conditions of diffraction. The Talbot effect is regarded as the underlying cause of pattern formation in optical cavities.
arXiv:2407.06312v2 Announce Type: replace-cross Abstract: Many systems resist analytical modeling, making data-driven inference of dynamics important. Yet data-driven methods can fail to converge or generalize, leaving open a central question: When can system behavior be learned reliably from data, and when is such learning impossible? We answer this question using adversarial dynamical systems to identify the boundary between accessible and inaccessible regimes. In Koopman operator learning, a leading framework for representing nonlinear dynamics through linear spectral objects, we design optimal data-driven spectral algorithms with convergence and certification guarantees under conditions arising broadly in physical systems. This yields a convergence theory for Koopman-operator approximations and resolves a longstanding open problem in Koopman spectral analysis. Conversely, by constructing adversarial systems, we prove matching impossibility results: without these conditions, no single-sequence limiting procedure can guarantee learning, regardless of data quality. These results sharply characterize when data-driven spectral learning can succeed and when it must fail. We validate the framework on oscillators, chaotic fluid flows and Arctic sea ice concentration forecasting. In the latter, we uncover hidden modes of Arctic sea ice decline, deliver long-range forecasts with geographic error bounds, and outperform state-of-the-art dynamical and deep learning models at substantially lower computational cost, enabling real-time deployment on standard CPUs.
arXiv:2508.00236v2 Announce Type: replace-cross Abstract: Interfaces govern reaction pathways and stability in electrochemical systems, yet creating clean, well-defined metal interfaces at scale remains challenging. In anode-free lithium metal batteries (AFLMBs), the current-collector interface is decisive for lithium nucleation and solid electrolyte interphase (SEI) formation, and ideally should support efficient charge transport, uniform reaction distribution, and long-term chemical and structural stability. Here we report an ion-implantation strategy that produces an atomically clean and oxidation-resistant copper interface. Implanting copper ions into commercial foils removes the native oxide while generating subsurface vacancy clusters directly beneath the surface -- an atomic-scale modification that does not increase the collector thickness but fundamentally alters interfacial chemistry. Experiments and multiscale simulations reveal that these vacancies act as strong oxygen traps, preventing reoxidation, enhancing interfacial conductivity, and guiding the formation of an ultrathin, Li2O-enriched SEI that promotes uniform lithium deposition and suppresses parasitic reactions. Applied in AFLMBs, the engineered current collectors deliver long-term stability with a Coulombic efficiency of 98.8% over 600 cycles under lean-electrolyte conditions. These findings demonstrate atomic-scale interface control of copper current collectors as a route toward stable and practical lithium metal batteries.
arXiv:2509.22879v2 Announce Type: replace-cross Abstract: Mixture models, such as Gaussian mixture models, are widely used in machine learning to represent complex data distributions. A key challenge, especially in high-dimensional settings, is to determine the mixture order and estimate the mixture parameters. We study the problem of approximating a target measure, available only through finitely many of its moments, by a mixture of distributions from a parametric family (e.g., Gaussian, exponential, Poisson), with approximation quality measured by the 2-Wasserstein or the total variation distance. Unlike many existing approaches, the parameter set is not assumed to be finite; it is modeled as a compact basic semi-algebraic set. We introduce a hierarchy of semidefinite relaxations with asymptotic convergence to the desired optimal value. In addition, when a certain rank condition is satisfied, the convergence is even finite and recovery of an optimal mixing measure is obtained. We also present an application to clustering, where our framework serves either as a stand-alone method or as a preprocessing step that yields both the number of clusters and strong initial parameter estimates, thereby accelerating convergence of standard (local) clustering algorithms.
arXiv:2511.01563v2 Announce Type: replace-cross Abstract: We determine the length of the shortest nontrivial geodesic loops on the Stiefel manifold endowed with any member of the one-parameter family of Riemannian metrics introduced by H\"uper et al. (2021). This family includes, in particular, the canonical and Euclidean metrics. By combining existing and new bounds on the sectional curvature, we determine the exact value of the injectivity radius of the Stiefel manifold under a wide range of members of the metric family.
arXiv:2512.06602v2 Announce Type: replace-cross Abstract: We develop a theoretical framework for high-harmonic generation (HHG) driven by quantum states of light based on a temporal-mode expansion of the electromagnetic field. This approach extends previous single plane-wave mode treatments to realistic pulse configurations and arbitrary multi-mode states of light, resolving conceptual inconsistencies arising from non-normalizable infinite plane waves and establishing consistency between analytical and numerical methods. We derive a correction factor that quantifies deviations from the diagonal approximation (in which the yield becomes a statistical average over classical-field simulations) both for the response of a single atom and in the many-atom regime. Our results confirms that the HHG spectrum for atoms driven by any quantum state of light in free space is accurately described by averaging semi-classical calculations over the Husimi distribution, with no observable genuine quantum effects in the spectrum. We also demonstrate that in the many-atom regime, the mean-field coherent-state approximation underlying this treatment does not preserve probabilities, although unitarity is restored by in the diagonal approximation. The absence of genuine quantum effects in the HHG yield is attributed to the large photon numbers ($\sim 10^{11}$) required to reach HHG intensities in free space, which render quantum fluctuations negligible. We discuss nanophotonic environments with ultrasmall mode volumes as potential platforms where few-photon strong-field processes could exhibit genuine quantum signatures.
arXiv:2606.23833v1 Announce Type: new Abstract: Terrestrial water storage (TWS) integrates snow, soil moisture, surface water, and groundwater and is a key indicator of how climate variability and human activity reshape the global water cycle. The GRACE and GRACE-FO satellite missions provide the only direct, globally consistent observations of TWS change, but their record only begins in 2002 which is too short for many climate-scale analyses. We present a deep learning application that reconstructs monthly GRACE-like TWS anomalies (TWSA) back to 1940 by learning the relationship between daily ERA5 meteorological forcing (precipitation, evapotranspiration, runoff) and monthly GRACE observations. In contrast to prior reconstruction approaches based on grid-cell-wise regression, CNNs, or LSTMs, we adapt a multi-variate time series graph neural network (MTGNN) architecture, which was originally developed for mobility and traffic forecasting on urban sensor networks to this satellite-geodesy task. Spatial dependencies are encoded in a static, interpretable hybrid adjacency matrix that combines geodesic proximity with lagged correlations of climatic time series, capturing both local hydrological coupling and large-scale teleconnections. The reconstruction achieves a grid-cell Pearson correlation of 0.69, a basin-mean correlation of 0.94, and a near-zero bias, and it reproduces the spatial fingerprints of the 2015/16 El Ni\~no and 2020/21 La Ni\~na events. A systematic comparison with established reconstruction approaches (GTWS-MLrec, RM-REC, GRAiCE) shows that the graph-based model is statistically competitive at basin scale, reaching a correlation within 0.025 of the best baseline while using only roughly half to a tenth of the predictors the other models require and revealing characteristic weaknesses in arid regions in all models. The complete implementation is publicly available at github.com/hcu-cml/MTGNN-TWS-Reconstruction-GRACE
arXiv:2603.04840v2 Announce Type: replace-cross Abstract: Speech production is a complex process spanning neural planning, motor control, muscle activation, and articulatory kinematics. While the acoustic speech signal is the most accessible product of the speech production act, it does not directly reveal its causal neurophysiological substrates. We present the first simultaneous acquisition of real-time (dynamic) MRI, EEG, and surface EMG, capturing several key aspects of the speech production chain: brain signals, muscle activations, and articulatory movements. This multimodal acquisition paradigm presents substantial technical challenges, including MRI-induced electromagnetic interference and myogenic artifacts. To mitigate these, we introduce an artifact suppression pipeline tailored to this tri-modal setting. Once fully developed, this framework is poised to offer an unprecedented window into speech neuroscience and insights leading to brain-computer interface advances. The source code and data are available.
arXiv:2604.22112v2 Announce Type: replace-cross Abstract: Tantalum is a promising platform for superconducting quantum circuits, yet coherence times remain limited by dielectric losses from interfacial two-level systems (TLS), exacerbated by native oxide regrowth. Here, we implement molecular surface passivation using self-assembled organic monolayers on freshly etched tantalum and silicon in coplanar waveguide resonators. Surface characterization by contact angle, XPS, FTIR and TEM confirm the formation of ordered, nanometer-thick films that suppress oxide formation. Microwave measurements in the ~5-9 GHz range reveal internal quality factors up to 1.8x10^6 in the single-photon regime at 100 mK, representing a ~140% improvement over untreated devices with native oxide. Power and temperature dependent measurements attribute this enhancement to reduced TLS-induced losses. These results demonstrate that molecular passivation effectively engineers low-loss interfaces and provides a scalable route toward high-coherence superconducting quantum devices.
arXiv:2605.02701v2 Announce Type: replace-cross Abstract: We propose a robust gradient estimator based on per-sample gradient clipping and analyze its properties both theoretically and empirically. We show that the resulting method, per-sample clipped SGD (PS-Clip-SGD), achieves optimal in-expectation convergence rates for non-convex optimization problems under heavy-tailed gradient noise. Moreover, we establish high-probability convergence guarantees that match the in-expectation rates up to polylogarithmic factors in the failure probability. We complement our theoretical results with multiple numerical experiments. In particular, we demonstrate that PS-Clip-SGD outperforms both vanilla SGD with momentum and standard gradient clipping when training AlexNet on the CIFAR-100 dataset, even after accounting for the additional computational time caused by per-sample clipping. We also empirically show that, in the presence of gradient accumulation, applying clipping at the mini-batch level can improve training performance while incurring virtually no additional computational cost. This finding is particularly interesting, as it contradicts the common practice of applying clipping only after all accumulation steps have been completed.
arXiv:2605.19208v2 Announce Type: replace-cross Abstract: Physical activity (PA) plays an important role in maintaining and improving health. Daily steps have been a key PA measure that is easily accessible with common wearable devices. However, methods are lacking to recommend a personalized optimal distribution of daily steps over a period of time for the best of certain health biomarkers. In this paper, we fill this void based on the data from the All of Us Research Program which includes months of step counts as well as repeated measurements of key health biomarkers. We develop a new offline reinforcement learning (RL) algorithm to learn personalized and optimal PA distributions associated with cardiometabolic risk, where the action is a function representing the daily step distribution over a period of time. Simulation studies demonstrate the advantage of the proposed approach over existing continuous-action RL methods. The learned optimal policy from the All of Us data generally suggests people take more daily steps and also follow a more consistent pattern of PA over time while offering tailored recommendations for subgroups in blood glucose level, body mass index, blood pressure, age, and sex.