Search arXiv⌕ Search

arXiv subjects

Bin Xu

Publications and source records attributed to Bin Xu.

At least 19 recordsLinked to original sources
arXiv2026

AB-Attention: Accurate Block Attention for Accelerating Video Diffusion Transformers

Video diffusion transformers generate high-quality videos, but the quadratic cost of 3D full attention dominates their inference. Dynamic block-sparse attention reduces this cost by computing exact attention only on key/value blocks selected from current activations. However, its fidelity degrades under aggressive sparsity: blocks are scored from coarse summaries or formed by costly global clustering, pruned blocks vanish from the softmax, and whole tiles are kept or dropped even when only a few of their keys matter. To address these limitations, we propose Accurate Block Attention (AB-Attention), a training-free method that accurately approximates dense attention at high sparsity. Its Intra-Block Clustering Selector (IBCS) estimates block importance faithfully yet cheaply by clustering queries only within each query block, without permuting tokens. Sampling with Inclusion-Probability Weighting (SIP) then samples blocks in proportion to these estimates and reweights them by inverse inclusion probability, so that one pass estimates dense attention rather than truncating it. Finally, its kernel gathers key/value blocks much smaller than a tile into full tensor-core tiles, refining selection at no extra tensor-core cost. On Wan2.1 and HunyuanVideo, AB-Attention skips 89--92\% of attention, whereas the baselines compute 1.8--3.6$\times$ as much, yet it still improves PSNR by up to 4.4~dB over the strongest baseline and achieves up to 2.23$\times$ end-to-end speedup. Our code is available at https://github.com/InfRec-Cardinal/AB-Attention.

cs.CV↗
arXiv2026

Local Models for Special Kähler Metric Singularities Along the Discriminant Locus of the $\mathrm{SL}_2(\mathbb{C})$ Hitchin Base

Freed (arXiv:hep-th/9712042) formulated the theory of special Kähler structures. In particular, the regular part of the $\mathrm{SL}_2(\mathbb{C})$ Hitchin base $\mathcal{B}$ carries such a structure, and the associated metric $ω_{\mathrm{SK}}$ develops singularities near the discriminant locus $\mathcal{D}\subset\mathcal{B}$. Baraglia--Huang (arXiv:1707.04975) showed that its Taylor expansion at regular points can be computed by topological recursion, and Hitchin (arXiv:1712.09928) later constructed integrable subsystems over smooth strata of $\mathcal{D}$ parametrizing spectral curves with only nodal singularities; these strata carry induced special Kähler structures. We show that near such strata, the special Kähler metric on $\mathcal{B}\setminus\mathcal{D}$ exhibits logarithmic asymptotics in transverse directions, whereas its tangential part converges to a metric on the stratum that agrees with the one induced by Hitchin's subsystems. We also prove that the induced metric on every radial complex line through the origin and a point of such a stratum is flat with cone angle $π$ at the origin.

math.DG↗
arXiv2026

REM: What to Store, When to Recall, What to Forget in Experience Memory for Coding Agents

Large language model (LLM) agents now autonomously complete long-horizon tasks, such as repository-level coding, through hundreds of think-act steps, yet the experience each run produces is rarely reused on the next. Prior memory systems concentrate on retrieval, finding the most relevant past trajectory. Retrieval cannot observe the running agent, so it cannot decide whether the agent needs an experience, at which point, or for how long. We introduce REM (Reusable Experience Memory), a framework that manages the memory lifecycle from inside the agent's think-act loop through three explicit decisions. What to store: each completed trajectory is consolidated into a phase-structured document, two orders of magnitude smaller than the trajectory. When to recall: the document is injected at the start of the task; the costlier implementation record is withheld until an in-loop gate detects the agent struggling. What to forget: the record is evicted after a single exposure. On repository-level bug fixing (SWE-ContextBench Lite), REM retrieves the related trajectory more accurately than every state-of-the-art memory system (68.7% vs. 55.6% match@1) and improves, in every replicate, end-task resolution (27.6% vs. 24.9%) at 17% lower cost. Injecting the same experience as a summary or raw trajectory, even when oracle-retrieved, does not. Ablating any of the three decisions reduces resolution. Altogether, these results demonstrate that the bottleneck in experience reuse is not retrieval fidelity but how experience is stored and governed across its lifecycle, opening a control-theoretic perspective on agent memory.

cs.AI↗
arXiv2026

Knowledge-Guided Construction of Machine Learning Potentials for Reactive Behavior in Energetic Materials

Energetic materials involve multiple interactions, making reliable machine-learning potentials (MLPs) difficult to construct without extensive datasets. Here, we enhance the ability of the neural network interatomic potential (NNAP) to describe molecular systems using sphmulti, a refined spherical-harmonic basis set that combines extra radial basis with element-specific chemical weighting in each radial channel. We further integrate hyperactive learning, farthest-point sampling, and iterative sufficiency criteria with knowledge-guided key events, enabling the NNAP to retain essential information for describing the material system. For 3,3'-diamino-4,4'-azoxyfurazan, the resulting NNAP is trained on only 13,680 configurations yet captures key properties, including the isomerization barrier, pyrolysis activation energy, and shock Hugoniot behavior. NNAP-driven molecular dynamics simulations further reveal that pyrolysis proceeds through furazan-ring opening followed by chain reactions, whereas shock loading induces anisotropic hydrogen transfer driven by interlayer structural deformations. This work enables efficient and reliable MLPs for energetic materials.

cond-mat.mtrl-sci↗
arXiv2026

Event-STVG: Spatio-Temporal Video Grounding for Long-Form Event Reasoning

Spatio-Temporal Video Grounding (STVG) aims to jointly localize a target in space and time given a natural-language query. Despite continued advances, existing STVG benchmarks remain largely centered on localizing targets through instance-specific descriptions, leaving them largely decoupled from long-form reasoning increasingly emphasized in multimodal large language models (MLLMs). In this paper, we introduce Event-STVG, a task that shifts STVG from locating what is described toward event-level reasoning by unifying long-form context, repeated-event reasoning, event localization, role inference, and multi-target grounding. To facilitate research on Event-STVG, we construct a large-scale benchmark comprising 4,016 annotated event occurrences across soccer and seven additional domains. Experiments with state-of-the-art MLLMs reveal substantial difficulty, with the strongest model achieving only 5.4% m-vIoU at the 10-minute horizon. Across domains, we consistently observe confusion among repeated same-type events, spatial grounding bottlenecks, and rare success in jointly grounding all required roles. Event-STVG establishes a new challenge for visual understanding in MLLMs, calling for future models that can jointly reason about complex events across space and time. Code and annotations are available at https://github.com/THU-KEG/EventSTVG

cs.CV↗
arXiv2026

Quantum-enhanced Search for Galactic Dark Photons Using Cat-State Cavities

We present a quantum-enhanced search for the Galactic dark-photon background (GDPB), a relativistic and anisotropic vector field that can be produced by dark-matter decay or annihilation in the Milky Way halo. We reinterpret an existing quantum-enhanced dark-photon dataset obtained with a superconducting circuit QED platform originally developed for wave-like dark-photon dark-matter searches, consisting of a high-$Q$ niobium cavity in the $\mathrm{TM}_{010}$ mode dispersively coupled to a transmon qubit. By preparing four-component cat states and performing repeated quantum non-demolition parity measurements, analyzed with a hidden Markov model, we extract photon-transition probabilities for multiple cat amplitudes and integration times. The coupling to the GDPB is governed by a polarization- and angle-dependent cavity overlap function, which predicts a distinctive diurnal modulation of the excitation rate for both longitudinal and transverse polarizations. From the absence of such modulation, we set 90\% confidence-level bounds on the kinetic-mixing parameter $ε$ for two benchmark production mechanisms (parametric and cascade decay) over the relativistic mass range $m_X \ll ω_c$.

hep-ph↗
arXiv2026

MERGED: Multimodal Entity Resolution via Generated Expert Reasoning Distillation

In product entity resolution, relationship definitions constantly evolve with business needs, yet adapting to each change traditionally requires slow, costly human annotation that is often noisy and carries no reasoning. Large vision-language models (VLMs) prompted zero-shot can adapt to a new definition immediately and supply the reasoning that human labels lack, but their cost and latency are prohibitive at production scale. We present MERGED, a distillation framework that transfers not just labels but structured reasoning from large teacher VLMs into a compact 7B-parameter student, requiring no human annotation for training. Multiple teachers label each product pair and articulate the reasoning behind their decision: agreement pairs supply supervised fine-tuning, while disagreements are resolved by a meta-judge into preference pairs for Direct Preference Optimization. Evaluated against human-labeled ground truth on a multilingual e-commerce dataset, the resulting student improves PR-AUC by 13.79% over the same backbone trained on human labels and surpasses the larger Qwen2.5-32B-VL baseline by 6.32% at 6x lower cost, while also yielding tighter label-reasoning consistency (over 10% above Qwen2.5-32B-VL). Moreover, re-applying MERGED from an existing checkpoint adapts to a new relationship definition with only 10K samples, improving PR-AUC by 6.97% over zero-shot and outperforming from-scratch training. MERGED enables rapid adaptation to evolving relationship definitions, supporting a new one in days rather than months, at a cost and latency suitable for large-scale industrial deployment.

cs.IR↗
arXiv2026

Observation of coherently modulated thermal phonon band and lifetime in superlattice

Manipulating phonon band structures through wave interference in artificial periodic nanostructures is of fundamental interest and forms the foundation of phonon band engineering. Yet, direct observation of thermal phonon wave interference has remained elusive despite half a century of extensive research. Here, utilizing inelastic x-ray scattering, we directly observed momentum-resolved coherently modulated thermal phonon bands and lifetimes in a short-period GaAs/AlAs superlattice at 300 K. The modulation persists at even 500 K, demonstrating robust coherent thermal phonon transport at high temperatures. Our measurements, combined with ab initio lattice-dynamics calculations, further show that thermal phonon interference expands the three-phonon scattering phase space and strengthens anharmonic effects, highlighting the interplay between phonon band modulation and anharmonicity. Our observations establish nanostructuring as a versatile route for engineering thermal phonon bands and lifetimes at high temperatures across diverse applications.

cond-mat.mtrl-sci↗
arXiv2026

Reduced Solutions of Toda Systems for Arbitrary Complex Simple Lie Algebras

We study reduced solutions of Toda systems generated by spherical metrics for an arbitrary complex simple Lie algebra. The main result is an intrinsic centralizer parametrization: for a spherical metric, the reduced family is the positive Iwasawa slice cut out by the centralizer of the principal image of closure of monodromy group. As an application, we transfer existing spherical-metric existence results to the diagonal Toda source tuple.

math-ph↗
arXiv2026

VUDA: Enabling Controlled Spatial Sharing of Graphics and Compute on NVIDIA GPUs

Graphics and compute increasingly share a single GPU in embodied AI simulators, AI-enabled games, and VR systems. Running these workloads concurrently can improve utilization, but contention can also compromise rendering and inference latency. Effective sharing therefore requires control over both concurrency and resource allocation. Native CUDA and Vulkan runtimes complicate this task: they place work in separate scheduling domains, while compute-oriented resource controls do not govern the graphics pipeline. We present VUDA, a system that enables controlled spatial sharing of native CUDA compute and Vulkan graphics on NVIDIA GPUs. An analysis of GPU scheduling, address translation, and workload dispatch reveals how to coordinate the two execution stacks without replacing either one. VUDA redirects CUDA channels into Vulkan's scheduling domain while preserving the runtimes' separate data address spaces. It then uses GPU front-end controls to partition execution resources between the graphics and compute pipelines, with allocations adjustable at runtime. Together, these mechanisms let applications control resource sharing without modifying GPU drivers or rewriting kernels and shaders. We evaluate VUDA across four application scenarios on three NVIDIA GPU platforms, covering throughput and latency objectives for graphics, compute, or both. In embodied AI simulation, enabling concurrency improves throughput by up to $1.62\times$ over the same asynchronous pipeline under default time sharing. In driving perception, partitioned co-execution reduces the measured inference-budget miss rate from 80.1% to zero while sustaining 60-FPS rendering.

cs.OS↗
arXiv2026

Polynomial area growth for special Kähler metrics

Let $g=w|dz|^2$ be an affine special Kähler metric on a punctured disc and let $Ξ=q(z)\,dz^3$ be its associated holomorphic cubic differential. We prove that if the $g$-area of concentric punctured neighbourhoods grows at most polynomially, then $q$ is meromorphic at the puncture. More precisely, an area bound of order $\eps^{-N}$ forces every pole of $q$ to have order strictly less than $N+3$. Hence an essential singularity of the cubic differential rules out every polynomial area upper bound. The proof associates to $(g,Ξ)$ a curvature $-1$ conformal pseudometric and applies the Ahlfors--Schwarz lemma, followed by the submean inequality for holomorphic functions. We also discuss the identically zero cubic differential and the borderline logarithmic growth responsible for the strict pole-order bound.

math.DG↗
arXiv2026

Reduction of Zieve's Conjecture on the Hurwitz problem to Three Branch Points

M. E. Zieve conjectured that a candidate datum over the sphere is realizable whenever the gcd of the parts of every branch partition is one and it satisfies the lcm condition. We prove that this conjecture reduces to its three-branch-point case. More precisely, if every Zieve-admissible triple is realizable, then every Zieve-admissible datum with at least four branch points is realizable. In particular, Zieve's conjecture implies the prime-degree conjecture posed by A. L. Edmonds, R. S. Kulkarni and R. E. Stong. Generative-AI disclosure. OpenAI's GPT-5.6 Sol and GPT-6 Astra assisted throughout the research and preparation of this article, including the exploration of reduction strategies, proof development and checking, consistency checks on notation and terminology, and manuscript drafting and revision. The authors reviewed all AI-assisted material with careful scrutiny and take full responsibility for all mathematical claims and for the final text.

math.DG↗
arXiv2026

DeepWeaver: Bridging the Evidence Synthesis Gap in Open-Ended Question Answering

Retrieve-then-generate pipelines are commonly used to produce deep-research answers for open-ended questions, but retrieval alone is insufficient: LLMs must organize noisy and fragmented evidence into comprehensive, well-cited answers. We refer to this process as evidence synthesis. However, direct generation often underuses evidence, misaligns citations, and collapses diverse information into shallow summaries, exposing an evidence synthesis gap between retrieval and generation. Thus, we propose DeepWeaver, a novel framework that weaves noisy retrieved evidence into comprehensive answers by maintaining Thought Block Chains (TBCs), a structured representation that groups claims, salient information, keywords, and supporting evidence. DeepWeaver uses subordinate TBCs to inspect residual evidence, commit TBC revisions, and discover new claims before final generation. We evaluate DeepWeaver on open-ended QA over both knowledge bases and the web, and introduce LoQA, a high-density benchmark for evidence synthesis. Across multiple LLMs, DeepWeaver improves content sufficiency, citation grounding, and detail preservation on LoQA, while achieving deeper insights and higher citation quality on DeepResearch Bench. These results show that evidence weaving is an effective mechanism for bridging retrieval and generation in open-ended QA. Our code is available at https://github.com/KlozeWang/DeepWeaver.

cs.CL↗
arXiv2026

CogEvol: Towards Efficient and Reliable Learning Environment Generation

We present CogEvol, a family of models trained specifically for Learning Environment Generation: turning a course brief into a finished learning artifact (structured-JSON slides or self-contained interactive HTML pages) in a single pass. Across 220k production requests, CogEvol completes a slide in a median of 17 seconds and an interactive page in 59, replacing minutes-long multi-turn agent scaffolding. Reliability is enforced rather than hoped for: a production-grounded data pipeline turns real failures into 53,687 verified SFT samples, and a hybrid rule-plus-VLM reward drives GRPO-based RL, hardened after we caught and fixed a reward-hacking episode that produced visually convincing but unplayable games. CogEvol-27B scores 83.7 on slide quality and 63.7 on a 500-case interactive-HTML benchmark with 26.9x fewer parameters than flagship coding models, and, in collaboration with the OpenMAIC team, serves their live production traffic. CogEvol-4B is released openly under the Apache 2.0 license at https://github.com/CogEvol/CogEvol-4B; external flagships are measured on the same suites under the identical harness. Scaffold editing cuts interactive-page generation cost by a further ~76%, and the full stack runs on domestic Ascend accelerators at application-level parity with A800 GPUs, lowering the unit cost of AI-native education at scale.

cs.CL↗
arXiv2026

Lightweight Machine Learning-Driven Monocular Sidewalk Path Extraction for Embedded Micromobility Navigation

Sidewalk-scale path extraction demands perception and planning that run reliably on compact, low-power hardware in cluttered, map-sparse environments. We present a monocular vision pipeline for sidewalk path extraction in micromobility systems that progresses through three design iterations, from a skeleton-graph baseline through distance-transform corridor planning to a lightweight image-space architecture, and provides a systematic comparison of five path-planning methods across both bird's-eye-view (BEV) and image-space domains. A compact SegFormer-B0 student model, trained with a semi-supervised teacher-student framework using OneFormer Swin-L pseudo-labels, achieves a hand-annotated IoU of 0.946 at 11.7 ms per frame, improving over the baseline checkpoint (IoU 0.758, 18.9 ms). In a controlled planner comparison on 32 hand-labeled frames, image-space midpoint planning achieves the lowest lateral center error (14.3 px) at 2.2 ms, a 421x speedup over BEV distance-transform planning (926.8 ms, 65.0 px center error), while maintaining comparable mask-path alignment (98.5% versus 98.6%). A full-video replay across six campus sequences (22,679 frames) confirms that the improved segmentation reduces temporal instability from 1.46% to 0.33% and increases template-path availability from 73.7% to 79.3%. We further show that BEV-only path extraction is fragile in monocular settings: in one profiled run, 99.3% of frames produced no valid BEV path. The final recommended architecture, image-space midpoint primary, image-space distance-transform fallback, and BEV reserved for visualization, runs the full perception-to-path stack in under 50 ms per frame on CPU, making it suitable for embedded pedestrian-speed micromobility systems.

cs.CV↗
arXiv2026

Zariski-Dense Monodromy of Singular Hyperbolic Metrics on Non-Hyperbolic Riemann Surfaces

We prove that the monodromy group of every singular hyperbolic metric on a non-hyperbolic Riemann surface, in the sense of potential theory, is Zariski dense in ${\rm PSL}(2,\mathbb{R})$, confirming a conjecture of the authors. The main new step is to show that a singular hyperbolic metric on an arbitrary parabolic Riemann surface cannot have monodromy contained in a conjugate of the real affine subgroup of ${\rm PSL}(2,\mathbb{R})$. The same argument also gives a direct proof in the compact case. Combined with the nonexistence results for the remaining proper subgroup types, this proves the conjecture.

math.DG↗
arXiv2026

Can We Defend Against AI-Generated Video Attacks on Real-World Crisis Events? A Systematic Evaluation of Detectors, Generators and Social Dissemination

Recent video generators can fabricate realistic depictions of wars, disasters, public emergencies, and other real-world crises, creating substantial risks of misinformation. Existing benchmarks, however, provide limited evidence on detector and generator behavior in such settings, including how detectability varies with generation conditions, how people perceive generated videos, and whether detectors remain reliable during social dissemination. To address this gap, we introduce RA-Bench, a benchmark for AI-generated video detection that uses Real videos as Anchors. RA-Bench contains 17,886 videos, comprising 1,830 real-video anchors across 10 social-risk categories and 16,056 generated clips from four open-source and five closed-source generators. Based on RA-Bench, we organize our evaluation along three dimensions. We first assess detector generalization across seven traditional detectors, ten zero-shot multimodal models under three review settings, and two MLLMs specifically fine-tuned on AI-generated video detection. Across these methods, none of the three detector families generalizes consistently across RA-Bench instances. We then examine how detectability varies with generation quality, conditioning information, and sampling seeds. These analyses show that generation properties affect detector families differently, while source-level detection patterns remain stable across seeds. Finally, we study human authenticity judgments and detector reliability during social dissemination. We find that videos that mislead people are also difficult for current detectors, and that social dissemination makes detection harder. Together, these findings show that current methods struggle to detect realistic AI-generated videos, highlighting the need for detectors robust to evolving video generators.

cs.CV↗
arXiv2026

Limit Values of Character Sums in Frobenius Formula of Three Permutations

We study the asymptotic behaviour of the character summation part of the Frobenius formula for three conjugacy classes of the symmetric groups. When two conjugacy classes contain fixed points of order $\sim H_i\sqrt{n}$ for $i=1,2$ and all other cycles are long, the summation converges uniformly to $2e^{-H_1H_2}$; the exponential factor coincides with the non-collision probability of the birthday paradox. When short cycles are rare in all three classes, character sum tends to $2$.

math.GM↗