arXiv subjects

Tao Xu

Publications and source records attributed to Tao Xu.

At least 19 recordsLinked to original sources

arXiv2026

OrDA: Orthogonal Disentanglement of Access Habits Framework for Homepage Marketing Block Recommendations

Clicks on homepage marketing blocks are driven by a dual-mechanism of content interest and access habits. However, habitual clicks often create Pseudo-Positives in marketing slots, where position advantage masks mediocre content quality, leading to biased recommendation ecosystems. We propose a framework called Orthogonal Disentanglement of Access habits (OrDA) to purify interest signals. OrDA utilizes a dual-tower structure with a gated allocation layer to adaptively route features and minimize interference. To ensure rigorous separation, we employ orthogonal regularization to constrain the latent interest and habit manifolds to be geometrically perpendicular. OrDA performs causal intervention (do-calculus) during inference to rank items solely by purified interest scores. Empirical online evaluations on large-scale datasets demonstrate that OrDA effectively eliminates access-habit bias, outperforming state-of-the-art methods in predictive accuracy. Online AB test 5.64% shows user click-through rates (UCTR) improvement on the Zhima homepage marketing block, Zhima rent-floor recommendation.

cs.LG↗

arXiv2026

A Physics-guided Fine-tuned LLM-based Framework for Customized Power Distribution System Feeder Generation

Power distribution system feeder models (e.g., IEEE 33-bus system, IEEE 13-bus system, etc.) are cornerstones for conducting power distribution system studies. As real-world feeder models are hard to acquire due to energy security concerns, generating high-quality synthetic feeders becomes an important alternative to satisfy the fast-growing and diversified needs of power system researchers and engineers. In this paper, we propose an LLM-based synthetic feeder generation framework that can achieve end-to-end generation from natural language specifications to physically consistent feeder models. First, Supervised Fine-Tuning (SFT) is performed on a dataset created following physical laws to empower the LLM with syntactic understanding of complex feeder structures. Second, Group Relative Policy Optimization (GRPO) with a specially-designed multi-stage gated reward function is introduced to better align the generation results with user intent and physical constraints. Third, a dual-agent architecture is deployed to refine and evaluate the generated feeders. Specifically, a refinement agent calibrates the feeder model parameters referring to the industrial feeder design standards, while a judge agent provides quality assessments. Case studies demonstrate that the proposed framework generates customizable feeders with valid formats, physical consistency and high engineering applicability.

eess.SY↗

arXiv2026

UniviewVLA: A Unified Multiview Vision-Language-Action Model with World Modeling

Occluded tasks remain a bottleneck in robot manipulation. Existing solutions either deploy additional physical cameras requiring training-inference camera parity, or rely on explicit 3D reconstruction with high computational cost. Moreover, both approaches rely on standard agent-view and wrist-view observations, while failing to capture occlusion information and future scene evolution. To this end, we propose UniviewVLA, a unified multiview Vision-Language-Action model with world modeling, which infers multiview scene evolution for action prediction from only standard two-camera observations. We demonstrate that by leveraging generated multiview future views from the world model, UniviewVLA reveals occluded cues and models future scene evolution, improving action prediction and removing the need for extra hardware or explicit reconstruction. Besides, to accelerate inference while preserving prediction accuracy, UniviewVLA develops Motion-Informative Token Compression, which compresses each generated view from 625 to 16 tokens and reduces per-view latency from 6-7s to 0.2-0.3s. UniviewVLA also proposes training-free Action-Entropy View Selection, which dynamically identifies the most action-informative view at different inference stages. Extensive experiments show that UniviewVLA achieves 95.8% on LIBERO and 4.60 on CALVIN ABCD to D, both standard occlusion-free benchmarks. On customized occlusion-focused tasks, it improves success rate from 40.0% to 73.3%, and average real-robot success rate by 33.4 points, demonstrating stronger occlusion-focused performance without sacrificing standard occlusion-free benchmarks.

cs.RO↗

arXiv2026

ATHENA: Accelerated Multi-Task Heterogeneous Influence Functions for Robot Data Curation

In robot imitation learning, influence functions provide a principled approach to quantify each demonstration's effect on robot task outcomes, yet scaling them to billion-parameter Vision-Language-Action (VLA) models is limited by computational and multitask bottlenecks. To this end, we propose ATHENA, an influence function framework tailored for multitask VLA data curation at a billion-parameter scale. Concretely, it leverages the Kronecker structure of linear-layer gradients to reduce projection cost, and approximates dense Hessian inversion with a rank-r Random Truncated Approximation, achieving about a 313.4x speedup in influence computation. Furthermore, ATHENA formulates global and local interactive influence to balance data curation across 50 jointly trained tasks. Extensive evaluations on RoboTwin 2.0 and real-robot deployment, covering 9.34 and 6.90 hours of demonstrations, respectively, show that ATHENA matches or exceeds full-data joint fine-tuning using only 50% of demonstrations in simulation and 66.7% of data across six real-robot tasks. Overall, ATHENA demonstrates its effectiveness for data curation in billion-parameter multitask VLA fine-tuning.

cs.RO↗

arXiv2026

VGPT-RSI for RH-Adjacent Formal Progress: Boundary Certificates, Verified Finite Lagarias Inequalities, and Explicit Failure Localization

The Riemann Hypothesis remains one of the central unsolved problems in mathematics. Rather than claiming proof, we investigate whether a verifiable AI-assisted reasoning system can produce reliable, formally checked partial progress while explicitly identifying the remaining mathematical obstructions. We apply the Verifiable Growing Physical Transformer with Recursive Self-Improvement (VGPT-RSI) to two RH-adjacent certification tasks. First, we construct and verify a finite RH-boundary certificate for inequality on a parameterized safe lower curve over a region. The numerical boundary curve is converted into a certificate-backed lower curve, audited using outward-rounded interval arithmetic and Arb/FLINT ball arithmetic, and then checked in Rocq/CoqInterval for the parameterized theorem. Second, we initiate a formal Lagarias-route certificate. Lagarias criterion states that RH is equivalent to the global inequality. We formalize the finite quantity and produce a Coq-checked finite certificate. The final system identifies the exact unresolved mathematical bottlenecks: formalizing the Lagarias equivalence, proving the global tail theorem beyond any finite cutoff, and potentially reducing counterexamples to colossally abundant or related extremal integers. These results demonstrate that VGPT-RSI can produce certified RH-adjacent formal progress, organize proof dependencies, and avoid overclaiming when the remaining obstruction is genuinely mathematical.

cs.AI↗

arXiv2026

Probing New Degrees of Freedom with the Universal Tail of Primordial Black Hole Mass Functions

The primordial black hole (PBH) mass function today develops a low-mass evaporation tail whose shape is universal with respect to the initial PBH mass distribution. This universality is fixed by the continuity equation and the Hawking mass-loss rate, but the tail is distorted if additional particle degrees of freedom participate in Hawking evaporation. Since this tail controls high-energy PBH photon emission, such distortions leave characteristic features in the gamma-ray spectrum. We show that these features provide a robust probe of new degrees of freedom, even for subdominant PBHs, within reach of future experiments and independent of visible-sector couplings or relic abundance.

hep-ph↗

arXiv2026

EntSQL: A Benchmark for Grounding Text-to-SQL in Long-Context Enterprise Knowledge

Text-to-SQL enables natural language access to databases, and recent LLMs have substantially advanced its capabilities. Existing benchmarks such as Spider, BIRD, and Spider~2.0 evaluate schema generalization, large-scale databases, and realistic workflows, but largely overlook enterprise scenarios where SQL generation depends on private business knowledge, such as internal metrics, reporting conventions, and organizational rules. We introduce EntSQL, an enterprise-oriented Text-to-SQL benchmark for evaluating long-context grounding over proprietary business documents. EntSQL contains 1,066 aligned Chinese-English semantic examples across five business domains, with most examples requiring domain knowledge beyond the question and schema and involving complex SQL structures. On English inputs, the best evaluated system reaches only 15.9\% when long-form documents are provided, highlighting the difficulty of grounding SQL generation in enterprise knowledge.

cs.CL↗

arXiv2026

Near-Optimal Mixed Strategy for Zero-Sum Linear-Quadratic Differential Games

Deriving analytic solutions for optimal mixed strategies in zero-sum linear-quadratic differential games (ZSLQDGs) remains an open problem. In this paper, we analytically synthesize near-optimal mixed strategies for ZSLQDGs and establish rigorous performance certifications. Specifically, we construct a surrogate pure-strategy stochastic differential game (SDG) by matching the first two moments of the mixed strategies. This method achieves an $\mathcal{O}(\bar{\pi}^2)$ weak approximation of state distributions and expected costs with respect to the maximum commitment delay $\bar{\pi}$. By analytically resolving the surrogate SDG, we derive closed-form optimal control laws for the matched moments. Crucially, we reveal that the surrogate game is governed by a Generalized Riccati Differential Equation (GRDE), which explicitly dictates a dynamic energy allocation law for variance injection. Building on these solutions, we propose a robust dual-routing architecture to execute the near-optimal mixed strategies. Furthermore, we certify that both the global value approximation error and the strategy suboptimality gaps are bounded by $\mathcal{O}(\bar{\pi}^{\frac{1}{2}})$. Finally, numerical experiments on a double-integrator pursuit-evasion game illustrate the induced physical behaviors and validate the theoretical bounds.

math.OC↗

arXiv2026

Near-Optimal Mixed Strategy for Zero-Sum Differential Games

Synthesizing near-optimal mixed strategies for zero-sum differential games (ZSDGs) has been a longstanding challenge. Existing research mainly focuses on characterizing the theoretical value function, while the practical design of executable mixed strategies remains open. To address this issue, we propose a novel weak approximation framework. The core idea is to map the original mixed-strategy game into a surrogate stochastic differential game (SDG) under pure strategies. This mapping ensures that both state distributions and cost expectations closely match the original game. Based on the solution of this auxiliary SDG, the original game value can be approximated, and near-optimal mixed strategies can be synthesized. To operationalize this framework, we develop a constructive control-space discretization algorithm for general ZSDGs. By parameterizing the infinite-dimensional measure optimization into standard probability simplices and solving local linear programs, our method efficiently synthesizes executable mixed strategies. Furthermore, we rigorously prove that the global weak approximation error is strictly of order $\mathcal{O}(\barπ)$ with respect to the maximum commitment delay $\barπ$, and derive explicit analytical upper bounds for the strategy suboptimality gaps. Numerical examples are provided to illustrate and validate our theoretical results.

math.OC↗

arXiv2026

Xiaomi OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation

Chain-of-Thought (CoT) reasoning has become a powerful driver of trajectory prediction in VLA-based autonomous driving, yet its autoregressive nature imposes a latency cost that is prohibitive for real-time deployment. Latent CoT methods attempt to close this gap by compressing reasoning into continuous hidden states, but consistently fall short of their explicit counterparts. We suggest that this is due to purely linguistic latent representations compressing a symbolic abstraction of the world, rather than the causal dynamics that actually govern driving. Thus, we present OneVL (One-step latent reasoning and planning with Vision-Language explanations), a unified VLA and World Model framework that routes reasoning through compact latent tokens supervised by dual auxiliary decoders. Alongside a language decoder that reconstructs text CoT, we introduce a visual world model decoder that predicts future-frame tokens, forcing the latent space to internalize the causal dynamics of road geometry, agent motion, and environmental change. A three-stage training pipeline progressively aligns these latents with trajectory, language, and visual objectives, ensuring stable joint optimization. In inference, the auxiliary decoders are discarded, and all latent tokens are prefilled in a single parallel pass, matching the speed of answer-only prediction. Across four benchmarks, OneVL becomes the first latent CoT method to surpass explicit CoT, delivering superior accuracy at answer-only latency. These results show that with world model supervision, latent CoT produces more generalizable representations than verbose token-by-token reasoning. Code has been open-sourced to the community. Project Page: https://xiaomi-embodied-intelligence.github.io/OneVL

cs.CV↗

arXiv2026

Liquid Crystal Theory of Biomembranes

Biomembranes, primarily composed of lipid bilayers, are not merely passive barriers but dynamic and complex materials whose shapes are governed by the principles of soft matter physics. This review explores the shape problem in biomembranes through the lens of material science and liquid crystal theory. Beginning with classical analogies to crystals and soap bubbles, it details the application of the Helfrich elastic model to explain the biconcave shape of red blood cells. The discussion extends to multi-layer systems, drawing parallels between the focal conic structures of smectic liquid crystals, the geometries of fullerenes and carbon nanotubes, and the reversible transitions in peptide assemblies. Furthermore, it examines icosahedral self-assemblies and shape formation in two-dimensional lipid monolayers at air/water interfaces. At the end of the paper, we find that the shapes such as cylinders, spheres, tori, biconcave discoids and Delaunay surfaces form a group. This result is merely an intrinsic geometric feature of these shapes and is independent of the biomembrane equation. When the pressure on the membrane, surface tension, and bending modules meet certain conditions, the biomembrane will take on these shapes. The review concludes by highlighting the unifying power of continuum elastic theories in describing a vast array of membrane morphologies across biological and synthetic systems.

physics.comp-ph↗

arXiv2026

Distributionally Robust Probabilistic Prediction for Stochastic Dynamical Systems

Probabilistic prediction of stochastic dynamical systems (SDSs) aims to accurately predict the conditional probability distributions of future states. However, accurate probabilistic predictions tightly hinge on accurate distributional information from a nominal model, which is hardly available in practice. To address this issue, we propose a novel functional-maximin-based distributionally robust probabilistic prediction (DRPP) framework. In this framework, one can design probabilistic predictors that have worst-case performance guarantees over a pre-defined ambiguity set of SDSs. Nevertheless, DRPP requires optimizing over the space of probability measures with density functions with respect to the Lebesgue measure, which is generally intractable. We develop a methodology that equivalently transforms the original maximin from function spaces to Euclidean spaces. Although it remains intractable to seek a global optimal solution, two suboptimal solutions are derived. By relaxing the constraints on the ambiguity set, we obtain a suboptimal predictor called Noise-DRPP. Relaxing the constraints on the predictor yields another suboptimal predictor, Eig-DRPP. Moreover, optimality gaps between the proposed predictors and the global optimal predictor are derived. Finally, we conduct elaborate numerical simulations to compare the performance of different predictors under different SDSs.

math.OC↗

arXiv2026

TCL: Enabling Fast and Efficient Cross-Hardware Tensor Program Optimization via Continual Learning

Deep learning (DL) compilers rely on cost models and auto-tuning to optimize tensor programs for target hardware. However, existing approaches depend on large offline datasets, incurring high collection costs and offering suboptimal transferability across platforms. In this paper, we introduce TCL, a novel efficient and transferable compiler framework for fast tensor program optimization across diverse hardware platforms to address these challenges. Specifically, TCL is built on three core enablers: (1) the RDU Sampler, a data-efficient active learning strategy that selects only 10% of tensor programs by jointly optimizing Representativeness, Diversity, and Uncertainty, substantially reducing data collection costs while maintaining near-original model accuracy; (2) a new Mamba-based cost model that efficiently captures long-range schedule dependencies while achieving a favorable trade-off between prediction accuracy and computational cost through reduced parameterization and lightweight sequence modeling; and (3) a continuous knowledge distillation framework that effectively and progressively transfers knowledge across multiple hardware platforms while avoiding the parameter explosion and data dependency issues typically caused by traditional multi-task learning. Extensive experiments validate the effectiveness of each individual enabler and the holistic TCL framework. When optimizing a range of mainstream DL models on both CPU and GPU platforms, TCL achieves, on average, 16.8x and 12.48x faster tuning time, and 1.20x and 1.13x lower inference latency, respectively, compared to Tenset-MLP.

cs.LG↗

arXiv2026

Direct Collapse Black Hole Candidates from Decaying Dark Matter

Injecting 1-13.6 eV photons into the early universe can suppress the molecular hydrogen abundance and alter the star formation history dramatically enough to produce direct collapse black holes. These, in turn, could explain the recently observed population of puzzling high-redshift supermassive black holes that appear to require super-Eddington accretion. We show that axion dark matter decay in the intergalactic medium can account for this energy injection. We use a single zone model of the gas core and semi-analytically evolve its chemo-thermal properties to track the conditions for which the system becomes an atomic cooling halo-a necessary precursor for the production of heavy black hole seeds to explain the high-redshift black hole population. Windows of axions masses between 24.5-26.5 eV with photon couplings as low as $4\times 10^{-12}$/GeV may realize this atomic cooling halo condition. We highlight the significance of the band structure of molecular hydrogen on the effectiveness of this process and discuss estimates of the heavy seed population and prospects for testing this model.

hep-ph↗

arXiv2026

Inverse scattering theory for the discrete PT-symmetric nonlocal nonlinear Schröinger equation under arbitrarily large nonzero boundary conditions

In this paper, the theory of inverse scattering transform (IST) is developed for the discrete PT-symmetric nonlocal nonlinear Schröinger equation under large nonzero boundary conditions (NZBCs). By considering that the data at infinity have constant amplitudes, two cases are studied where the previous IST theory fails for large NZBCs. Based on a suitable uniformization variable, the rigorous proofs for the analyticity, symmetries and asymptotic behaviors of the eigenfunctions and scattering coefficients are provided for the direc problem, and the potential reconstruction formula is derived by solving the Riemann-Hilbert problem. Particularly, the focusing equation is found to admit two types of novel solitons under large NZBCs: oscillating soliton and breather, where the former has not been previously reported, while the latter does not occur under small NZBCs. In addition, the multi-soliton solutions are shown to exhibit the collisions among oscillating dark/anti-dark solitons, and the superposition of oscillating soliton and breather.

nlin.SI↗

arXiv2026

Huawei Cloud Model-as-a-Service on the CloudMatrix384 SuperPod

Scaled-out MoE LLMs and scaled-up SuperPods create new systems challenges for production Model-as-a-Service (MaaS), requiring disaggregation, low-latency communication, and decentralized serving. This report presents xDeepServe, the production serving system behind Huawei Cloud's MaaS offering on CloudMatrix384, a 48-server SuperPod with 384 Ascend 910C chips connected by a high-bandwidth UB fabric and global shared memory. It serves models including DeepSeek, Kimi, GLM, Qwen, and MiniMax, among others. xDeepServe is built around Transformerless, a disaggregated execution architecture that decomposes transformer inference into modular units -- attention, feedforward, and MoE -- and supports disaggregated Prefill-Decode and MoE-Attention deployments. To enable disaggregation, we develop XCCL, a memory-semantic communication layer providing microsecond-level point-to-point and scalable all-to-all primitives, and we extend FlowServe with decentralized DP groups and techniques to mitigate stragglers and synchronization variance. In a peak decoding configuration, xDeepServe reaches 2400 tokens/s per Ascend 910C chip at ~50ms time-per-output-token (TPOT).

cs.DC↗

arXiv2026

Index Light, Reason Deep: Deferred Visual Ingestion for Visual-Dense Document Question Answering

Existing multimodal document question answering methods predominantly adopt a Pre-Ingestion (PI) strategy: during the indexing phase, a Vision Language Model (VLM) is called on every page to generate page descriptions that are then encoded into vectors, and questions are answered via embedding similarity retrieval. However, this approach faces a dual dilemma on visual-dense engineering documents: VLM blind descriptions inevitably lose critical visual details, and embedding retrieval systematically fails on highly similar documents. This paper proposes the Deferred Visual Ingestion (DVI) framework: zero VLM calls during preprocessing, leveraging only document structural information (table of contents, drawing numbers) to automatically build a hierarchical index through the HDNC (Hierarchical Drawing Number Clustering) algorithm; during inference, candidate pages are located via BM25 retrieval, and the original images along with the specific question are sent to a VLM for targeted analysis. Large-scale experiments on three datasets validate the effectiveness of DVI: on Bridge engineering drawings (1,323 questions), end-to-end QA accuracy reaches 65.6\% vs. PI's 24.3\% (+41.3pp); on Steel catalog (186 questions), 30.6\% vs. 16.1\% (+14.5pp); on CircuitVQA, a public benchmark (9,315 questions), retrieval ImgR@3 achieves 31.2\% vs. 0.7\%. On the Bridge dataset, we evaluated ColPali (ICLR 2025 visual retrieval SOTA), which achieved only 20.1\% PageR@3, demonstrating that the failure of embedding retrieval on homogeneous engineering documents is structural rather than due to insufficient model capability. Ablation studies show that HDNC zero-cost automatic indexing yields a +27.5pp retrieval improvement, and VLM conversion rate analysis confirms that the bottleneck lies on the retrieval side rather than the comprehension side.

cs.CL↗

arXiv2026

AI-driven Large-scale Electron Microscopy enables Whole-tissue Subcellular Digitization

The distribution and interactions of cellular organelles play a critical role in mediating cellular physiology and pathology. Large-scale electron microscopy enables visualization of organelle distribution and interactions at the tissue level with nanometer resolution, but robust and efficient computational analysis tools are lacking. Here, we present a deep learning tool for universal large-scale 2D/3D electron microscopy analysis, DeepOrganelle. This new tool enables high-throughput, cell-resolved spatiotemporal mapping and digitization of organelle distribution and interactions. When applied to spermatogenesis across 12 stages and 22 differentiation status of the germ cells, DeepOrganelle uncovered previously unrecognized, stage-dependent dynamics of mitochondria-endoplasmic reticulum contact sites within one subphase of prophase I during meiosis. It also revealed coordinated organelle redistribution in Sertoli cells towards the blood-testis barrier, digitizing the remodeling dynamics of the tissue. This study demonstrates that DeepOrganelle provides a powerful framework that captures subcellular dynamics at the whole-tissue level.

physics.bio-ph↗