arXiv subjects

Jiawei Li

Publications and source records attributed to Jiawei Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

Soft Curriculum Learning for Optimizing Fresh and Generalized Recommendations

Large-scale recommender systems, particularly short-form video platforms, are often bottlenecked by massive popularity feedback loops. In such environments, as models recommend popular items, they generate an overwhelming amount of skewed training data for "head" items. This creates a self-reinforcing cycle where retrieval and ranking models memorize "head" item patterns at the expense of generalizing across the vast "tail" of the catalogue. While Curriculum Learning (CL) offers a powerful mechanism to break this feedback loop by systematically exposing models to progressively more difficult and less frequent examples, its adoption in industrial recommendation has been hampered by hardware utilization inefficiencies or the needs for complicated pre-processing techniques because dynamic data rejection algorithms tend to starve hardware accelearators (TPUs/GPUs) by becoming largely CPU-bound. In this work, we introduce a scalable Soft Curriculum Learning framework designed specifically for continuous training setups within industry-scale retrieval and ranking models. By utilizing loss annealing and in-graph weight adjustments rather than rigid data filtering, we break the popularity feedback and enable dynamic curriculum pacing without sacrificing system throughput. We demonstrate empirical evidence through applications across sequence-based retrieval models (such as SASRec), two-tower retrieval models, and large-scale continuous ranking models. Online A/B tests on our short-video platform demonstrate substantial lifts in both overall user satisfaction and fresh content consumption, all without degrading model throughput.

cs.IR↗

arXiv2026

GDLAM: Group-Disentangled Latent Action Model for Highly Disentangled Embodied Pretraining

Latent action models (LAMs) learn action-related representations from action-free videos via self-supervised future prediction, offering a scalable paradigm for embodied intelligence pretraining. However, existing LAMs collapse heterogeneous sources of visual change, including camera motion, object dynamics, and interaction events, into a single latent vector, resulting in entangled representations with limited semantic structure and consequently restricting world model controllability and VLA policy generalization. We introduce the Group-Disentangled Latent Action Model (GDLAM), a latent action model whose code is factorized by construction into N groups, each with an independent variational bottleneck and a spatially gated routing pathway, and trained with a set of information-geometric objectives: mutual exclusivity, group and gate sparsity, and static-dynamic orthogonality, that make the groups mutually causally distinct rather than merely decorrelated. Quantitatively, intervening on any single group changes only that group and leaves the others intact, and GDLAM improves label-free disentanglement metrics, including Modularity, MIG, and DCI, by wide margins over a strong unstructured LAM. Notably, this factorization is not at the expense of action information: across three mutual-information estimators and a linear probe, the grouped code is more informative than monolithic baselines both in- and out-of-distribution. As supporting evidence that the disentangled code is a reusable pretraining currency, we further transfer it to two downstream regimes: (1) World Modeling: World models pretrained with GDLAM achieve superior rollout fidelity and action-following capability compared with SOTA baselines. (2) VLA Policies: Pretraining with GDLAM substantially improves task success rates over previous methods across multiple simulation benchmarks and real-world robotic manipulation tasks

cs.RO↗

arXiv2026

LAWM-3D: Learning 3D-Aware Latent Actions from Human Videos for Generalizable Robot World Models

World models enable agents to perform forward rollout and planning without real-world interaction. However, their application in open-world embodied intelligence remains limited by the high cost of action annotations and the heterogeneity of action spaces across platforms. Recently, latent action models (LAMs) have alleviated this bottleneck by learning action representations directly from unlabeled human videos in a self-supervised manner. Nevertheless, most existing LAMs rely on single-view inputs and operate primarily in 2D pixel space, raising a fundamental question: can simply incorporating multi-view videos into LAM training endow the learned latent actions with 3D-aware perception? Our study shows that the answer is negative. The primary reasons lie in future-frame appearance leakage as well as inter-camera appearance discrepancies and viewpoint variations. To address these issues, we propose LAWM-3D, which introduces three tightly coupled key designs: (1) a multi-view invariant unified action tokenization scheme for learning 3D-aware latent actions; (2) a geometric alignment constraint that anchors intermediate encoder features to a pretrained 3D foundation model, thereby explicitly providing cross-view geometric correspondences; and (3) a non-injective RGB-D joint reconstruction objective that prevents shortcut learning from future-frame appearance information, forcing the LAM to focus supervision on motion cues with geometric significance. Importantly, these components are not simply stacked but are tightly coupled through a unified motivation. Built upon a two-stage paradigm of large-scale human video pretraining followed by robot fine-tuning, extensive experiments demonstrate that the proposed 3D-aware latent actions significantly improve world model performance, achieving SOTA results in generation quality, physical consistency, and generalization ability.

cs.CV↗

arXiv2026

Beyond Flat Policies: Hierarchical Post-Training for Embodied Agents in Robotic Manipulation

Vision-language-action (VLA) models have demonstrated remarkable capabilities in robotic manipulation by leveraging pretrained vision-language models. However, existing post-training methods predominantly optimize VLA models as flat policies, making it difficult to explicitly model task progression and perform robust long-horizon manipulation. Although hierarchical approaches introduce task decomposition, they mainly rely on supervised learning from offline demonstrations and cannot effectively improve execution through online interaction. To address this limitation, we propose Hierarchical Robotic Control (HiRoC), a hierarchical post-training framework that decouples high-level task planning from low-level action execution. The planner decomposes complex tasks into executable subgoals to provide explicit semantic guidance, while the executor continuously improves subgoal-conditioned action generation through reinforcement learning. To enable effective collaboration between the two modules, we further align the executor with planner-generated subgoals before reinforcement learning, mitigating the distribution misalignment between planning and execution. Extensive experiments across diverse robotic manipulation benchmarks demonstrate that HiRoC consistently outperforms strong baselines. Comprehensive analyses further validate the effectiveness of hierarchical post-training and the contribution of each key component.

cs.RO↗

arXiv2026

Enhancing LLM Performance Through Debate: An Empirical Study on Multi-Agent Debate for Coding Tasks

Large Language Models (LLMs) have advanced autonomous agents' planning and decision-making, yet they struggle with complex tasks requiring diverse expertise and multi-step reasoning. Multi-Agent Debate (MAD) systems, introduced in NLP research, address this gap by enabling structured debates among LLM-based agents to refine solutions iteratively. MAD promotes divergent thinking through role-specific agents, dynamic interactions, and structured decision-making. Recognizing parallels between Software Engineering (SE) and collaborative human problem-solving, this study investigates MAD's effectiveness on four coding tasks in SE. We adapt a MAD framework from NLP, analyze agent interactions to assess consensus-building and iterative refinement, and propose two MAD variants that enhance agent debate for coding tasks by addressing the observed weaknesses. Our findings show that structured debate and collaboration improve problem-solving and yield strong performance in some cases, highlighting the collaborative debate synergy between LLM agents for coding tasks in SE while identifying areas for future exploration.

cs.SE↗

arXiv2026

Geometric-Topological Perception and Motion Prior for Real-Time Satellite Video Object Tracking

Satellite video object tracking (SVOT) remains fundamentally challenging due to texture scarcity, arbitrary rotation, aspect ratio changes, and severe occlusions. While recent state-of-the-art trackers excel in general scenarios, their reliance on rich appearance details or rigid spatial matching mechanisms leads to significant performance degradation in the satellite domain. To bridge this gap, we propose SiamGM, a real-time spatial-temporal unified tracking framework tailored for satellite videos. Instead of conventionally stacking modules, we synergize geometric-topological perception with temporal-kinematic prior, addressing the core limitations of SVOT. Spatially, we closely couple a Topological Attention Module (TAM) with a Geometry-Constrained Label Assignment (GCLA) method during the training phase, where a Local Graph Propagation (LGP) mechanism enforces neighborhood-consistent template-search correspondences. This joint paradigm establishes fine-grained structural correspondences and explicitly suppresses surrounding background noise. Temporally, rather than passively relying on visual cues, we propose an Online Motion Model Refinement (OMMR) strategy during the tracking phase. Gated by an online self-calibrated confidence indicator, OMMR adaptively fuse long-term stable and short-term instantaneous motion estimation for robust tracking. Evaluations on two challenging SatSOT and SV248S benchmarks confirm that SiamGM outperforms not only recent Siamese-based satellite trackers but also state-of-the-art Transformer trackers in both precision and success metrics. Notably, this highly unified architecture introduces only minor computational overhead, enabling real-time tracking at 130 frames per second (FPS).

cs.CV↗

arXiv2026

ROCS: Request-Oriented Compute Sharing for Efficient Large-Scale Recommendation

Modern recommendation models gain prediction quality by scaling feature-interaction and sequence modules, but production cost constraints cap how far systems can scale. In this work, we propose Request-Oriented Compute Sharing (ROCS), a modeling and inference paradigm that exploits a unique property of recommendation inference: each user request is evaluated against many candidates, while request-side features are shared across candidates. ROCS defers request-candidate interactions as late as possible, isolates candidate-dependent representations, and evaluates substantial portions of the model once per request rather than once per candidate, significantly improving inference efficiency while maintaining or improving prediction quality. To realize this paradigm, we develop Generalized Layer Masking (GLM) to enforce candidate isolation in feature-interaction architectures, and Deep Cross Attention (DCA) to extend request-oriented sharing to sequence architectures. To support efficient GPU deployment, we co-design In-Kernel Broadcast Optimization (IKBO) that significantly accelerates ROCS model execution. Experiments on public benchmarks show that ROCS consistently improves the quality-efficiency tradeoff across recommendation backbones. On production-scale workloads, ROCS achieves up to a 3x QPS improvement on retrieval models without quality degradation and a 0.5% relative LogLoss improvement with a 50% QPS gain on a short-form video ranking model. ROCS has been deployed across large-scale recommendation systems spanning ads and organic surfaces, retrieval and ranking stages, and more than two orders of magnitude in inference complexity, delivering significant online gains at reduced infrastructure cost.

cs.LG↗

arXiv2026

Haicheng and Tangshan Earthquakes as potential Dragon-Kings

The dragon-king earthquake hypothesis proposes that some very large to great earthquakes are not merely the extreme end of the frequency-magnitude Gutenberg-Richter distribution (FMD), but are generated by distinct physical mechanisms, making them statistical outliers. We develop a data-driven framework to systematically test the dragon-king earthquake hypothesis. Our method combines objective spatial clustering, based on data-adaptive kernel density estimation (KDE), with a high-power sequential outlier detection technique. For each identified cluster, we exam sine the tail of the FMD to identify anomalous events. Candidate dragon-kings are evaluated via robust statistical tests, primarily the max-robust-sum (MRS) test with inward sequential testing. For each observed statistic of the MRS test, we calculate its p-value defined as the probability that this statistic could be generated by the null distribution. We apply this framework to seismicity surrounding the 1975 Haicheng mL 7.4 and 1976 Tangshan mL 7.9 earthquakes. For Haicheng, the mainshock shows a strong dragon-king signature in its pre-mainshock sequence, with p-values between 0.03 and 0.07 across a stable range of KDE density thresholds used to define natural seismicity clusters. Post-mainshock and combined sequences yield slightly higher p-values (up to 0.09). In contrast, the Tangshan mainshock exhibits a weaker outlier signal before the event (p-values in the range 0.05-0.15) but a stronger dragon-king signature afterward, with p-values from 0.015 to 0.05. The evidence that the Haicheng and Tangshan mainshocks exhibit dragon-king characteristics supports the idea that some large and great earthquakes arise from a maturation process that may enhance predictability. Haicheng provides a clear case study, while retrospective analyses suggest that Tangshan might also have been forecasted under more systematic procedures.

physics.geo-ph↗

arXiv2026

Hermes - Towards an Optimal High-Performance Algorithm for Cosmic Statistics of Large Data Sets

We present Hermes, an in situ multiresolution framework for efficient and flexible measurements of cosmic large-scale-structure statistics from discrete catalogues. Hermes reconstructs a catalogue as a continuous density field in a compact scaling-function basis and replaces explicit counting of particle tuples with algebraic operations among window-filtered fields. Standard binning schemes for counts-in-cells, two-point and higher-order correlation functions are thereby expressed through choices of window functions, while new statistics can be constructed by modifying the kernels without redesigning the estimator. We introduce PyHermes, an open-source Python implementation combining multiresolution reconstruction, FFT-based convolution, MPI/thread parallelism, and GPU acceleration. It supports isotropic and anisotropic two-point statistics, marked correlations, standard and multipole three-point functions, filtered statistics, and differential operators for derived physical fields. Tests with cosmological N-body halo catalogues demonstrate a range of clustering measurements and quantify the computational efficiency and scalability of the approach. By separating field representation from statistical windows, a single reconstructed field can be reused for many standard and customised measurements, making Hermes well suited to large data sets from current and future galaxy surveys.

astro-ph.CO↗

arXiv2026

AgentOmnia: Scaling Agentic Models for Full-Scenario Applications

Large language model agents have advanced rapidly, yet progress remains fragmented across domains, capabilities, task difficulty, and interaction settings. We frame this as full-scenario agentic scaling and present AgentOmnia, a framework coordinating task-space definition, data synthesis, post-training, evaluation, and improvement across To-Consumer (ToC), To-Business (ToB), and To-Employee (ToE) applications. An extensible Domain x Capability x Atomic Difficulty taxonomy aligns these stages and enables fine-grained diagnosis with OmniaBench. AgentOmnia combines bidirectional environment-task synthesis with tool-dependency, program-structured, and solver-based pipelines, constructing 5,018 stateful environments with 255,375 tools and 52,361 tasks. Programs, solvers, and verifiers provide correctness signals, while supervised fine-tuning, online agentic reinforcement learning, and a rollback curriculum support post-training. Evaluation failures translate into Product Requirement Documents (PRDs) for targeted self-evolution. Starting from Qwen3-30B-A3B-Thinking-2507, AgentOmnia raises the pass rate on the OmniaBench challenging subset from 9.16% to 37.11% and the macro-average across OmniaBench, $τ^2$-Bench, DeepPlanning, and VitaBench from 22.86% to 41.69%. Under a unified protocol,it leads the evaluated agentic post-trained baselines on OmniaBench and retains the highest four-benchmark macro-average. It also surpasses Qwen3-235B-A22B-Thinking-2507 on all four benchmarks and exceeds Qwen3.5-35B-A3B on the macro-average. Gains span three application splits, ten capability dimensions, eight atomic-difficulty factors, and 76 of 90 level-1 domains, indicating broad rather than category-specific improvement. A one-round study provides initial evidence for PRD-guided self-evolution, motivating validation at larger scales and in industrial settings.

cs.AI↗

arXiv2026

Decoupling Cross-Modality Manifold Discrepancy: Leveraging Visible Diffusion Priors for Infrared Super-Resolution

Infrared image super-resolution (IISR) mitigates the limitations imposed by low spatial resolution. Existing methods have recognized that IISR should preserve consistency in global distribution and structural information while enhancing image clarity. However, these methods are either insufficient or overly intrusive, a problem that becomes even more pronounced in diffusion-based models. To address these issues, we propose a dual-path diffusion-based framework for IISR, termed Shift-IISR. The proposed method is designed to improve the consistency of IISR results while preserving the generative capacity of diffusion models. Specifically, we develop a Global Representation Modulation (GRM) module to extract modality-specific information from infrared imagery and guide the global distribution of the diffusion model toward the ground truth. In addition, we introduce a Local Structure Refinement (LSR) module to encourage the model to focus on structural information at each step of the iterative denoising process. Extensive experiments demonstrate that the proposed method effectively improves distributional and structural consistency while maintaining competitive super-resolution performance. The source code of the proposed Shift-IISR can be available at https://github.com/Assassink8/Shift-IISR.

cs.CV↗

arXiv2026

Avoiding Obfuscation with Prover-Estimator Debate

Training powerful AI systems to exhibit desired behaviors hinges on the ability to provide accurate human supervision on increasingly complex tasks. A promising approach to this problem is to amplify human judgement by leveraging the power of two competing AIs in a debate about the correct solution to a given problem. Prior theoretical work has provided a complexity-theoretic formalization of AI debate, and posed the problem of designing protocols for AI debate that guarantee the correctness of human judgements for as complex a class of problems as possible. Recursive debates, in which debaters decompose a complex problem into simpler subproblems, hold promise for growing the class of problems that can be accurately judged in a debate. However, existing protocols for recursive debate run into the obfuscated arguments problem: a dishonest debater can use a computationally efficient strategy that forces an honest opponent to solve a computationally intractable problem to win. We mitigate this problem with a new recursive debate protocol that, under certain stability assumptions, ensures that an honest debater can win with a strategy requiring computational efficiency comparable to their opponent.

cs.AI↗

arXiv2026

Unified Deflection Estimation and Error Analysis for Background-Oriented Schlieren

Background-Oriented Schlieren (BOS) has become a versatile quantitative diagnostic for density-varying flows, in which estimating the light-ray deflection from the measured displacement is the essential step linking the recorded images to the underlying refractive-index field. Two-dimensional BOS traditionally treats this through the intuitive deflection angle, whereas three-dimensional tomographic BOS relies on the rigorous deflection vector derived from the ray equation. These descriptions have evolved largely independently, and the assumptions bridging them, together with the systematic errors they introduce, have not been examined in a unified manner. Based on geometric optics, this study establishes a unified deflection estimation framework that reconciles the mainstream two- and three-dimensional methods into a single mathematical structure and exposes the hierarchy of approximations underlying each. By deconstructing four key assumptions, namely the thin phase object, the uniform boundary refractive index, the paraxial approximation, and the perpendicularity between the deflection vector and the optical axis, we derive rigorous unified deflection expressions in both two- and three-dimensional space and categorize the mainstream methods accordingly. Using phase objects constructed from one-dimensional chirp signals and two-dimensional turbulent fields from Direct Numerical Simulation, combined with high-fidelity nonlinear ray tracing as the ground truth, we quantitatively characterize and analytically interpret the deflection estimation error of each method under both uniform and non-uniform refractive-index boundary conditions. This work provides a theoretical toolkit for assessing and enhancing the accuracy of quantitative BOS diagnostics.

physics.flu-dyn↗

arXiv2026

OmniaBench: Benchmarking General AI Agents Across Diverse Scenarios

Large language models are increasingly evolving from text generators into general agents capable of understanding user requests, invoking external tools, and completing complex tasks through interaction. However, existing agent benchmarks often focus on limited scenarios, tool ecosystems, or interaction formats, making it difficult to systematically characterize model capabilities across heterogeneous application settings. We introduce OmniaBench, a benchmark for evaluating general agents across diverse scenarios with explicit state spaces. We derive application-oriented scenario knowledge from app stores, product documents, industry resources, Web retrieval, and human refinement, forming a hierarchical taxonomy that spans ToC, ToB and ToE with 90 level-1 and 354 level-2 domains. Based on this taxonomy, we construct executable environments and synthesize single-turn and multi-turn tasks through four complementary routes: DAG, DAG-S, Solver, and Program. OmniaBench further introduces a ten-dimensional capability taxonomy and eight compositional atomic difficulty factors to support fine-grained evaluation and analysis. The resulting dataset contains 1,431 tasks, together with a challenging subset of 644 tasks designed to reduce evaluation cost and mitigate potential contamination of the full set after public release. The bench presents substantial challenges to current frontier models, with even Claude-Sonnet-5 and GPT-5.6-Sol achieving Overall Pass@1 scores of only 58.54 and 57.14, respectively. Further analyses reveal clear differences across domains and capabilities, as well as persistent limitations in planning, constraint maintenance, and adaptive correction. OmniaBench provides a broad and diagnostic benchmark for characterizing the capability boundaries of general agents.

cs.CL↗

arXiv2026

JoyAI-Sim: A Simulation-Enabled Interconversion Toolchain for the Embodied Data Pyramid

Generalist robot policies require trustworthy evaluation and robot-usable training data, but both are difficult to scale with physical robots alone. Real-robot trials and demonstrations remain the most faithful source of deployment signals, yet they are slow, costly, and hard to reproduce. We present JoyAI-Sim, a simulation-enabled interconversion toolchain for human-robot aligned model evaluation and data generation, denoted as Robot $\rightleftharpoons$ Simulation $\rightleftharpoons$ Human. On the one hand, the Robot $\rightarrow$ Simulation $\rightarrow$ Human pathway supports human-robot aligned model evaluation by reconstructing real-robot tabletop organization tasks as calibrated digital twins for scalable evaluation, while using human embodied feedback to inspect and refine the naturalness of simulated motions. On the other hand, the Human $\rightarrow$ Simulation $\rightarrow$ Robot pathway supports human-robot aligned data generation: it lifts ego-centric human demonstrations into simulation, checks them under robot physical constraints, and converts them into robot-centered trajectories, annotations, and visual observations. Together, these pathways use the JoySim simulator as both a scalable evaluation layer and a physical consistency filter for robot data generation. We further package the core reconstruction, simulation, rendering, and realism-augmentation modules as cloud services on JD Cloud, turning the system into a reusable and scalable infrastructure for robot data generation and model evaluation.

cs.RO↗

arXiv2026

A remark on pathwise well-posedness of the 1-$d$ stochastic heat equation

We study pathwise well-posedness of the stochastic heat equation (SHE) with a multiplicative noise on the circle. By combining the convolution Young and rough integration theory, introduced by Gubinelli and Tindel (2010), with the random tensor estimate approach to pathwise well-posedness of stochastic dispersive PDEs with multiplicative noises, introduced by Chapouto and the second and third authors (2026), we establish pathwise well-posedness of SHE in both the Young and rough cases, improving the results in Gubinelli and Tindel (2010). In particular, in the rough case (= the white-in-time case), our result covers the case of almost space-time white noise, thus establishing an optimal result within the framework of one-parameter rough paths.

math.AP↗

arXiv2026

Nonlinear PDEs with modulated dispersion II: Korteweg-de Vries equation

(Due to the limit on the number of characters for an abstract set by arXiv, the full abstract can not be displayed here. See the abstract in the paper.) We study dispersive equations with a time non-homogeneous modulation acting on the linear dispersion term. As primary models, we consider the Korteweg-de Vries equation (KdV) and related equations such as the Benjamin-Ono equation (BO) and the intermediate long wave equation (ILW), imposing certain irregularity conditions on the time non-homogeneous modulation. In this work, we establish phenomena called regularization by noise in three-folds: (i) When the modulation is sufficiently irregular, we show that the modulated KdV on both the circle and the real line is locally well-posed in the regime where the (unmodulated) KdV equation is known to be ill-posed. In particular, given any $s \in \mathbb R$, we show that the modulated KdV on the circle with a sufficiently irregular modulation is locally well-posed in $H^s(\mathbb T)$. Moreover, by adapting the $I$-method to the current modulated setting, we prove global well-posedness of the modulated KdV in negative Sobolev spaces. (ii) It is known that certain (semilinear) dispersive equations such as BO and ILW exhibit quasilinear nature. We show that sufficiently irregular modulations make the modulated versions of these equations semilinear by establishing their local well-posedness by a contraction argument, providing local Lipschitz continuity of the solution map. (iii) We also prove nonlinear smoothing for these modulated equations, where we show that a gain of regularity of the nonlinear part becomes (arbitrarily) larger for more irregular modulations. As applications of our approach, we also include further examples.

math.AP↗

arXiv2026

Nonlinear PDEs with modulated dispersion III: multiplicative noises

We investigate pathwise well-posedness of the stochastic modulated Korteweg-de Vries equation (KdV) on the circle with a multiplicative noise, where a time non-homogeneous modulation acts on the linear dispersion term. (i) In the Young case (= fractional-in-time case with Hurst parameter greater than $\frac 12$), we establish a new regularization-by-noise phenomenon on the stochastic convolution in a pathwise manner, where a gain of spatial regularity becomes (arbitrarily) larger for more irregular modulations. We then prove that, given any $s \in \mathbb R$ and any multiplicative Young noise, however rough it is in space, the stochastic modulated KdV is pathwise locally well-posed in $H^s(\mathbb T)$, provided that the modulation is sufficiently irregular. (ii) In the rough case (= white-in-time case), irregularity of the modulation does not induce any smoothing on the stochastic convolution, and in fact, there is a slight loss in the spatial regularity. In this case, by slightly regularizing the multiplicative noise term, we prove pathwise local well-posedness in $H^s(\mathbb T)$ for any given $s \in \mathbb R$, provided that the noise is sufficiently smooth in space. We achieve these goals by combining (i) the sewing lemma approach to the nonlinear Young integration theory, introduced by Chouk and the second author (2014), and (ii) the pathwise construction of stochastic convolutions as Young or rough integrals via the random tensor estimate and the sewing lemma, introduced by the first, fourth, and fifth authors (2026). In the appendix, we also present an example of regularization by noise for a stochastic modulated Schrödinger equation with a multiplicative Young noise.

math.AP↗