Searcharxiv⌕ Search

arXiv subjects

Bin Wang

Publications and source records attributed to Bin Wang.

At least 19 recordsLinked to original sources
arXiv2026

Clapping: Removing Per-sample Storage for Pipeline Parallel Distributed Optimization with Communication Compression

Pipeline-parallel distributed optimization is essential for large-scale machine learning but is challenged by significant communication overhead from transmitting high-dimensional activations and gradients between workers. Existing approaches often depend on impractical unbiased gradient assumptions or incur sample-size memory overhead. This paper introduces Clapping, a Communication compression algorithm with LAzy samPling for Pipeline-parallel learnING. Clapping adopts a lazy sampling strategy that reuses data samples across steps, breaking sample-wise memory barrier and supporting convergence in few-epoch or online regimes. Clapping comprises two variants including Clapping-FC and Clapping-FU, both of which achieve convergence without unbiased assumption for compressed gradient, effectively addressing compression error propagation in multi-worker settings. Numerical experiments validate the performance of Clapping across different tasks.

math.OC↗
arXiv2026

Anthropogenic Regional Adaptation in Multimodal Vision-Language Model

While the field of vision-language (VL) has achieved remarkable success in integrating visual and textual information across multiple languages and domains, there is still no dedicated framework for assessing human-centric alignment in vision-language systems. We offer two contributions to address this gap. First, we introduce Anthropogenic Regional Adaptation: a novel paradigm that aims to optimize model relevance to specific regional contexts while ensuring the retention of global generalization capabilities. Second, we present a simple, but effective adaptation method named Geographical-generalization-made-easy (GG-EZ), which utilizes regional data filtering and model merging. Through comprehensive experiments on 3 VL architectures: large vision-language models, text-to-image diffusion models, and vision-language embedding models, and a case study in Southeast Asia (SEA) regional adaptation, we demonstrate the importance of Anthropogenic Regional Adaptation and the effectiveness of GG-EZ, showing 5-15% gains in cultural relevance metrics across SEA while maintaining over 98% of global performance and even occasionally surpassing it. Our findings establish Anthropogenic Regional Alignment as a foundational paradigm towards applicability of multimodal vision-language models in diverse regions and demonstrate a simple-yet-effective baseline method that optimizes regional value alignment while preserving global generalization.

cs.AI↗
arXiv2026

Quantized valley source in bilayer graphene via adiabatic quantum pumping

We propose a quantized valley source based on adiabatic quantum pumping in bilayer graphene in the presence of both quantum spin Hall and quantum valley Hall effects. By periodically modulating the gate voltages of two quantum dots coupled to a bilayer graphene nanoribbon, the helical edge states enable the generation of a pure valley current. Depending on the geometry of quantum dots, the number of emitted valley pairs either forms quantized plateaus or varies continuously with the pumping amplitude. The quantized plateau structure can be tuned by the static gate voltage and by the coupling between the quantum dots and the ribbon. Quantized valley pairs are robust against Anderson disorder applied in the quantum dots and persist for both zigzag and armchair quantum-dot boundaries. These results demonstrate that adiabatic quantum pumping provides a route toward electrically controllable quantized valley sources.

cond-mat.mes-hall↗
arXiv2026

Hierarchy of Electromagnetic Gauge Field Quantum Geometries

The conventional Bloch quantum geometric tensor (QGT) describes electric-dipole geometry for spatially uniform perturbations and vertical Bloch transitions. We develop a broader framework linking quantum geometry to local and nonlocal electromagnetic response in gauge-field space, placing this tensor within a hierarchy of electric and magnetic multipole quantum geometries. The intrinsic antisymmetric adiabatic response in every electromagnetic channel is a Berry curvature in the space of its conjugate fields. The field-dependent projector uniquely fixes the QGT for that channel, pairing the response curvature with a quantum metric. This construction extends to spatially nonuniform perturbations, whose finite momentum transfer induces nonvertical Bloch transitions. We explicitly construct the current--current QGT connecting Bloch states at $\mathbf{k}\pm\mathbf{q}/2$, providing a unified metric and curvature description of nonlocal current response. In the doubled Brillouin zone, tangential derivatives generate conventional momentum-space QGT multipoles, whereas normal derivatives connect finite-momentum geometry to uniform electric- and magnetic-multipole-pair QGTs. We derive this connection through second order in $\mathbf{q}$. The finite-momentum curvature governs the intrinsic nonlocal Hall response, while an inverse-frequency moment of symmetric optical absorption probes the index-symmetric metric sector.

cond-mat.mes-hall↗
arXiv2026

Event2Reward: Learning Robotic Rewards from Progress Relations

Learning task progress from robot demonstrations provides a practical way to construct dense shaping rewards for robotic reinforcement learning. A common approach uses normalized demonstration time as progress labels, implicitly assuming monotonically increasing task progress. During policy exploration, however, actual task progress can stall or even regress. Such models may mistake regressions for progress, assigning positive shaping rewards to actions that cause progress loss. Real suboptimal and failed executions offer corrective experience, but assigning absolute progress values to their states is difficult. Our key observation is that stalls, regressions, and recoveries establish local relative progress relations even when absolute progress is unknown. We introduce Event2Reward, which uses normalized temporal labels to anchor the progress scale and these within-trajectory relations to constrain local changes, combining absolute and relative progress supervision. Its progress predictions serve as potentials for reward shaping. We curate an event-annotated dataset from approximately 100K raw real-robot execution records, covering successful, suboptimal, and failed executions, without requiring framewise absolute progress labels. Under the same training budgets, Event2Reward improves average policy success by 18.16 percentage points over a baseline using only sparse task rewards across 40 tasks in four LIBERO suites. On real robots, average success after 30 minutes of wall-clock training reaches 60.0% on three precise insertion tasks and 90.0% on three object manipulation tasks, exceeding the sparse-reward baseline under the same budget by 36.7 and 23.3 percentage points, respectively. Ablation studies further show that absolute progress supervision and relative progress relations are complementary, with joint supervision yielding better downstream policy learning.

cs.RO↗
arXiv2026

Compatibility of Perverse Truncation of Ngô Fibrations

We study compatibility of perverse truncations for morphisms between Ngô fibrations with rationally smooth total spaces. We show that, under an effective rank inequality, equivariant pullbacks preserve the normalized perverse truncations. Applied to the diagonal, this gives multiplicativity of perverse truncations for Hitchin systems with rationally smooth total spaces.

math.AG↗
arXiv2026

Rimuru: Unified Memory Management for Modern LLM Serving via Elastic Expert Offloading

Modern LLMs strain device memory through large model weights and attention states, particularly in resource-constrained scenarios. Existing systems separately provision memory for model weights and attention states, imposing fixed fixed boundaries that cannot adapt to the changing memory demands of workloads. Under heavy workloads, insufficient state capacity limits request concurrency and reduces throughput; under light workloads, idle memory in one region cannot be reallocated to another, wasting device memory. Meanwhile, modern LLM architectures introduce heterogeneous tensors with diverse layouts and lifetimes, further complicating memory management. These problems motivate unified memory management for heterogeneous tensors. However, a shared pool alone does not make memory directly reusable across tensors with different allocation granularities and execution layouts. This paper presents Rimuru, a modern LLM serving system that unifies memory management through elastic expert offloading. Rimuru establishes a shared page pool to enables elastic memory sharing across heterogeneous tensors. To accommodate different allocation granularities, Rimuru introduces memory views that allow each tensor type to interpret the shared memory space using its own page size and page table. Furthermore, Rimuru combines huge-page alignment with a multi-level page allocator that splits and merges pages, enabling cross-type capacity reuse. A page-table-aware fused MoE kernel directly accesses non-contiguous expert weights, avoiding copy overhead. Building on these mechanisms, Rimuru employs a workload-aware adjuster that dynamically reallocates pages across tensor types according to runtime demands. Implemented on top of vLLM, Rimuru remains compatible with modern LLMs. Evaluations show that Rimuru achieves up to 19.03x baseline throughput and reduces TTFT by 98.1% on average.

cs.OS↗
arXiv2026

A novel class of high-order uniformly accurate exponential integrators with local linear extension for the charged-particle dynamics under strong magnetic field

In this paper, we develop a novel class of high-order uniformly accurate exponential integrators for charged-particle dynamics under strong magnetic field. The small parameter $0<\varepsilon\ll 1$ induces rapid temporal oscillations, rendering traditional numerical methods prohibitively expensive due to severe step-size restrictions. To address this issue, a linearization technology that introduces auxiliary polynomial variables is employed to recast the original charged-particle dynamics as a higher-dimensional system. Classical exponential integrators are subsequently applied to this augmented formulation, which inherently carries richer structural information, thereby yielding a family of uniformly accurate exponential integrators that can reach arbitrarily high order without requiring any order conditions. For strong magnetic fields under the maximal-ordering scaling, we rigorously establish through algebraic techniques that the schemes with auxiliary polynomial variables of degree $k\geq2$ achieve an improved position error bound of $\mathcal{O}(\min\{\varepsilon^{k+1}h,\varepsilon h^{k+1}\})$ and a uniform velocity error bound of $\mathcal{O}(\min\{\varepsilon^{k}h,h^{k+1}\})$, where $h$ denotes the time-step size. Under the additional condition that the magnetic field gradient vanishes at the origin, these bounds sharpen to $\mathcal{O}(\min\{\varepsilon^{k+2},\varepsilon h^{k+1}\})$ and $\mathcal{O}(\min\{\varepsilon^{k+1},h^{k+1}\})$ for the position and velocity, respectively, for $k\geq1$. The same sharper bounds hold for constant strong magnetic fields. Numerical experiments confirm the advantages of the proposed methods.

math.NA↗
arXiv2026

Interior curvature estimates for convex solutions of a prescribed curvature quotient equation

In this note, we prove a priori interior curvature bounds for convex solutions to an elliptic Weingarten curvature quotient equation. The proof avoids integral estimates, doubling inequalities and compactness arguments. It relies on a concavity inequality for the equation operator and a novel choice of the auxiliary function to carry out an elementary maximum-principle argument. As an application, we obtain interior curvature bounds for convex solutions of a prescribed special Lagrangian curvature equation in dimensions three and four.

math.AP↗
arXiv2026

NxN E-valuation: Hypothesis Certification via a Conformal CRT Null

Language models now propose hypotheses faster than anyone can check them, so the step that certifies them must need nothing from the proposer but the hypothesis. We state three requirements for such a step -- per-hypothesis automatable, attributable and diagnostic -- and introduce NxN e-valuation, which meets them. A hypothesis names a scope, the units it is about, and gives each of them a restriction that scores its outcome. The units of one scope serve as nulls for one another, so no resampling model is written per hypothesis; the algorithm certifies a stable, scope-level association and reports whether the scope, the restriction, or which unit receives which restriction carries the prediction. For independently drawn units, betting e-values bound the chance of a false certificate for each hypothesis in finite samples, under stated nulls. A betting lower bound on the share of units that carry the association gates every certificate, and an asymptotically valid equivalence test calls the restrictions shared only when it rejects differences between them above a declared noise ceiling. We test it on planted data, on randomized intervention data, and on real purchase data with 285 hypotheses that language models wrote without knowing our method or our classes, labeled by human experts. Meeting all three requirements costs no accuracy: NxN makes 692 correct decisions in 693 with one false certificate, against 659 with four for a composite of CRT tests that returns our verdicts but needs a resampling model per hypothesis, and 224 with 468 for held-out validation. Removing any of our statistics merges a verdict that certifies with one that rejects. The certificates concern predictive association, not causality.

cs.AI↗
arXiv2026

From AI-assisted to AI-native Industrial Control: Architecture, Agent Design, and Challenges

The integration of AI and industrial control is rapidly transforming industrial automation. However, current approaches remain predominantly AI-assisted, where AI functions are incorporated as auxiliary components without fundamentally altering the underlying architecture of industrial control systems. Consequently, the potential of AI to enable higher levels of autonomy, adaptability, and continuous evolution remains underutilized. This paper proposes an AI-native industrial control paradigm, which reconceptualizes industrial control systems by positioning AI as the intrinsic core rather than an external enhancement. The proposed paradigm is built upon an AI-native infrastructure and a collection of collaborative agentic AIs, each encapsulating a set of advanced skills. Among them, the industrial control agent plays a central role in orchestrating the system lifecycle. Through this paradigm, industrial control shifts from a closed, capability-oriented, human-driven, and passive engineering process to an open, task-oriented, intent-driven, and autonomous solution delivery mechanism. The paper further analyzes the key challenges of the AI-native paradigm from both AI and industrial control perspectives, including issues related to AI model reliability, formal verification, real-time assurance, and heterogeneous resource orchestration. Overall, this work outlines a forward-looking vision for industrial control in the AI era and identifies critical research directions toward autonomous, evolvable, and adaptive control systems in complex and dynamic industrial environments.

physics.soc-ph↗
arXiv2026

An explicit third-order low-regularity integrator for the cubic nonlinear Schrödinger equation

In this paper, we propose an explicit third-order low-regularity integrator for the cubic nonlinear Schrödinger equation on the one-dimensional torus $\mathbb{T}$. By carefully designing numerical approximations for the integrals in Fourier space, we construct a third-order scheme and provide a rigorous global error analysis. More precisely, we prove that the scheme achieves third-order convergence in $H^{r}(\mathbb{T})$ for the initial data in $H^{r+4}(\mathbb{T})$ with $r>1/2$. Furthermore, the third-order convergence in $L^{2}(\mathbb{T})$ is established for the initial data in $H^{4}(\mathbb{T})$. Numerical experiments are presented to confirm the theoretical convergence results.

math.NA↗
arXiv2026

When Fusion Order Matters: Insights on Progressive Multimodal Representation Learning for Cross-modal Retrieval

Multimodal video retrieval combines visual, acoustic, and speech information, yet how their fusion order should depend on the data and the retrieval query remains unclear. We introduce Unified Progressive Representation Learning for Order Analysis, or UniPRO, to investigate this question through configurable fusion schedules with matched model components and training settings. Starting from text-to-video retrieval, we find that the tested one-by-one schedules outperform stacked fusion, while changing the order within one-by-one fusion also changes performance. To guide order selection, we examine two complementary considerations: learning difficulty under unequal modality availability and preservation of information relevant to the query. Complete-modality controls show that missing inputs alone cannot explain order sensitivity. At R@1, text and audio retrieval favor schedules that introduce the modality matching the query last. Controlled speech missingness during training and testing reverses the preferred order in both R@1 and SumR, with different crossover points. These findings motivate balancing the two considerations when selecting a fusion path. We then use them to predict a visual-last schedule for image queries, which achieves 97.9 R@1, the highest among the evaluated orders on MSR-VTT-9k. Main experiments on four text-to-video benchmarks establish the effectiveness of UniPRO, with component ablations assessing its supporting design. Together, the results identify fusion order as a task-dependent choice that should account jointly for modality availability and the retrieval query.

cs.AI↗
arXiv2026

AlphaAnything: Extending Pretrained RGB Generators to RGBA

Existing RGBA generation methods commonly learn a transparency-specific representation through reconstruction, introducing an additional training stage and potentially altering the RGB decoding function inherited from pretrained models. We present AlphaAnything, a lightweight framework that extends pretrained RGB generators to RGBA directly in their original latent space, without RGBA reconstruction pretraining. AlphaAnything inserts LoRA modules into the generative transformer and attaches an alpha branch to the frozen RGB VAE decoder. Unidirectional RGB-to-alpha feature sharing allows opacity supervision to shape the generative latent while guaranteeing that, for any fixed latent, RGB decoding remains identical to that of the pretrained decoder. We further introduce Endpoint-Aware Alpha Loss (EA Loss), which derives opacity supervision from the complementary foreground--background weights used in compositing. It supports exact opacity endpoints and continuous partial transparency while emphasizing localized alpha errors. Experiments on text-to-RGBA and image-to-RGBA tasks across multiple pretrained RGB backbones demonstrate competitive generation quality, and controlled ablations validate the shared-latent training design and EA Loss.

cs.CV↗
arXiv2026

Sci-CoE: Co-evolving Scientific Reasoning LLMs via Geometric Consensus with Sparse Supervision

Large language models (LLMs) have demonstrated exceptional reasoning capabilities, and co-evolving paradigms have shown promising results in domains with verifiable reward signals, such as code and math. However, in unverifiable domains, such as scientific reasoning, these methods remain fragile due to unreliable solution evaluation and limited diversity in verification strategies. In this work, we propose Sci-CoE, a two-stage co-evolving framework for domains without verifiable feedback. The framework enables models to self-evolve as both solvers and verifiers by transitioning from sparse supervision to unsupervised learning. In the first stage, the model uses a small set of annotated data to establish fundamental correctness judgment anchors for the Verifier. In the second stage, we introduce a geometric reward mechanism that jointly considers consensus, reliability, and diversity, driving large-scale self-iteration on unlabeled data. Experiments on several general scientific benchmarks demonstrate that Sci-CoE enhances complex reasoning capabilities and exhibits strong scalability, facilitating the construction of more robust and diverse evaluation systems. Codes are available at https://github.com/InternScience/Sci-CoE.

cs.AI↗
arXiv2026

AtomicRL: Atomic Reinforcement Learning for Long-Horizon Robotic Manipulation

Vision-language-action (VLA) models have shown promise in robotic manipulation, but imitation learning (IL) alone often struggles to master complex long-horizon skills. Reinforcement learning (RL) offers an effective avenue for further improving VLA policies. However, existing VLA-RL methods typically suffer from delayed and biased credit assignment due to trajectory-level sparse rewards, while imbalanced subtask mastery creates bottlenecks that limit overall performance. To address these challenges, we propose AtomicRL, an RL post-training framework that leverages atomic task structures for reliable credit assignment and adaptive exploration. AtomicRL introduces Semantic-Aware Policy Optimization (SAPO), which uses semantic-aware gating to constrain TD residuals from failed atomic subtasks from propagating across task boundaries, mitigating credit contamination while preserving intra-subtask credit assignment and value bootstrapping. In addition, Bottleneck-Adaptive Target Exploration (BATE) dynamically coordinates task sampling and local exploration incentives based on the mastery of individual atomic subtasks, directing learning toward the remaining bottlenecks. Finally, we develop a Real2Sim2Real pipeline that bridges online RL in simulation with real-world deployment, facilitating practical transfer to physical robots. On LIBERO-LONG, AtomicRL achieves absolute performance gains of 3.6% and 4.0% over PPO across two VLA architectures. On CALVIN ABC-D, it improves full-task performance by up to 9.5% over PPO. Combined with the Real2Sim2Real pipeline, AtomicRL further improves real-world task performance by an average of 21.3%. These results demonstrate that AtomicRL effectively improves the learning efficiency and manipulation performance of VLA policies on long-horizon tasks.

cs.RO↗
arXiv2026

DocLayout-YOLO: Axis-Attentive Document Layout Detector with Mesh-Guided Synthetic Pretraining

Accurate and efficient document layout analysis is critical for modern document parsing systems. However, although the YOLO framework represents a leading paradigm for the real-time detector, its adaptation to heterogeneous documents remains underexplored. To address this gap, we propose DocLayout-YOLO, a document-specialized real-time detector that integrates diverse synthetic document pretraining and page-axis context modeling. From the data perspective, we introduce MeshLayout-1M, a synthetic pretraining corpus generated by the proposed Mesh-candidate BestFit algorithm, which provides large-scale and diverse document layouts for pretraining and improves adaptation to heterogeneous document scenarios. From the model perspective, we propose Page-Axis Layout Attention (PALA), which captures long-range dependencies along the horizontal and vertical page axes while remaining computationally efficient. Extensive experiments demonstrate that DocLayout-YOLO achieves state-of-the-art performance on four heterogeneous document benchmarks, obtaining 72.4%/75.5%/96.0%/73.1% mAP50:95 on D4LA, M6Doc, HJDataset, and PRIMA-LAD, respectively. Compared with existing real-time detectors, DocLayout-YOLO establishes a superior accuracy-efficiency trade-off in document layout analysis. Code, model, and data are available at https://github.com/opendatalab/DocLayout-YOLO/tree/release-v2.

cs.CV↗