Search arXiv⌕ Search

arXiv subjects

Lei Zhang

Publications and source records attributed to Lei Zhang.

At least 37 records · Page 2Linked to original sources
arXiv2026

BitMTP: When Multi-Token Prediction Meets Low-Bit Large Language Models

For efficient deployment of large language models (LLMs), methods of ultra-low-bit backbones reduce weight storage, memory traffic, and linear-layer execution cost, while multi-token prediction (MTP) methods reduce sequential decoding steps by predicting multiple future tokens in parallel. Although complementary, the direct combination of these methods cannot accelerate decoding at comparable quality: masked drafting lacks intermediate token input supplied in next-token training, while wider computation and causal key-value (KV) updates can offset the savings from fewer steps. To address these challenges, we propose BitMTP, an algorithm-system co-design framework for efficient MTP on ultra-low-bit LLM backbones. A joint next-token prediction and MTP objective equips a single model with both autoregressive (AR) generation and block-causal drafting. Confidence-only decoding, speculative-only decoding, and hybrid decoding provide different quality-throughput operating points without using separate draft and verifier models. Accepted-prefix alignment keeps the committed prefix, materialized causal KV boundary, and CUDA Graph replay state consistent for continued generation. BitMTP uses this contract to select width-specialized low-bit execution and reuse captured graphs over shared KV storage. On NVIDIA H20, BitMTP with hybrid decoding achieves five-task mean speedups of 2.2x on BitNet b1.58-2B and 2.1-2.4x on 1.7B-4B Bonsai models at comparable task quality. On RTX 3090 and Jetson Orin, it further achieves up to 2.0x in five-task mean throughput and 3.3x on MBPP+, respectively. These results demonstrate that BitMTP accelerates ultra-low-bit LLM generation across backbones, model sizes, and GPU types. The source code is released at https://github.com/PolyU-VCLab/BitMTP

cs.CL↗
arXiv2026

Qwen-Audio-3.1-ASR Technical Report

We present Qwen-Audio-3.1-ASR, a family of audio-language models that builds upon its predecesor Qwen-Audio-3.0-ASR with improved speech recognition and expanded capabilities in speaker-diarized transcription and general audio understanding. The family retains support for 30 languages and 16 Chinese dialects, together with instruction-controlled capabilities including historical-context conditioning, hierarchical hotword customization, and native transcription polishing. Speaker-diarized ASR produces structured outputs containing speaker labels, timestamps, and recognized text, maintaining consistent speaker labels across turns and the representation of overlapping speech. The general audio understanding capabilities cover audio-, speech-, and music-captioning, temporal grounding and sound event detection, as well as audio question answering and multiple-choice reasoning. These tasks broaden the application scope from recognizing spoken words to interpreting speakers, acoustic events, and their temporal relations. To accommodate different deployment requirements, the family includes variants for asynchronous file transcription, low-latency streaming, interactive voice input, and non-streaming recognition, with speaker-diarized transcription and general audio understanding available in certain variants. Evaluations span conventional ASR, multilingual and dialectal speech, speaker-diarized transcription, and general audio understanding, demonstrating the capabilities of the Qwen-Audio-3.1-ASR family across transcription and broader audio analysis tasks.

cs.SD↗
arXiv2026

Benchmarking the Limits of In-Context Reinforcement Learning for Ad-Hoc Teamwork

In-Context Reinforcement Learning (ICRL) has enabled foundation agents to adapt instantaneously to novel tasks, yet its efficacy in Ad-Hoc Teamwork (AHT)-where coordination with unknown partners is required-remains unexplored. To rigorously evaluate this, we introduce a large-scale benchmark ICRL4AHT, built upon a high-throughput JAX implementation of Overcooked-V2. Our benchmark includes a large, diverse teammate suite spanning both RL and heuristic policies, enabling controlled train-test shifts, and provides a reproducible end-to-end pipeline for teammate generation, learning-history collection, dataset construction, and online multi-episode evaluation. We evaluate representative history-conditioned ICRL algorithms, including Algorithm Distillation (AD) and Decision-Pretrained Transformer (DPT), across millions of transitions. Results reveal notable limitations: contrary to their success in single-agent domains, these baselines fail to exhibit robust test-time adaptation in multi-agent settings. Specifically, these methods frequently underperform random baselines across both unseen teammate and unseen layout tracks, with no clear in-context improvement over long horizons. These findings highlight the challenges of strategic inference under partial observability within the OvercookedV2 AHT protocol, establishing our benchmark as a critical testbed for next-generation coordination algorithms.

cs.AI↗
arXiv2026

HarnessIR: Harnessing Multimodal Foundation Models for Universal Real-World Image Restoration

Real-world low-quality images suffer from complex mixed degradations, including but not limited to noise, blur, atmospheric effects, etc. Recent agentic methods usually model real-world image restoration (Real-IR) as a sequential tool calling problem over task-specific single-degradation restoration models. This paradigm, however, is fundamentally limited because complex real-world degradations cannot be cleanly undone degradation by degradation, and the tool used for task-specific models caps the capability of the agent system. In this work, we present HarnessIR, an agentic framework for Real-IR by harnessing a multimodal foundation model (MFM) as the executor. HarnessIR consists of five stages: perception and diagnosis, on-demand tool invocation, prompt composition, execution, and verification-driven refinement. Unlike prior agentic Real-IR methods that rely on tool chains assembled from task-specific models, HarnessIR feeds the restoration requirements, the perceptual diagnosis, and the evidence into an MFM that performs restoration in a single pass, followed by verification stages to determine whether the result warrants further processing. Under our harness, off-the-shelf MFMs handle restoration tasks remarkably well, achieving state-of-the-art results on the widely used MiO100 synthetic benchmark. More importantly, by exploiting the strong generalization ability of MFMs, HarnessIR delivers compelling restoration quality on challenging real-world scenes where previous agentic IR systems often struggle. Codes is available at https://github.com/PolyU-VCLab/HarnessIR.

cs.CV↗
arXiv2026

Magnetoelectric Bulk Photovoltaic Effect in Centrosymmetric Solids

It is commonly believed that a bulk photovoltaic current requires broken spatial inversion symmetry. However, in this work, we demonstrate that dc photocurrents can arise in centrosymmetric solids from the bilinear coupling between the electric and magnetic components of light. This magnetoelectric bulk photovoltaic effect (MBPE) has a quantum-geometric origin in the dipole moments of the Zeeman quantum metric and Zeeman Berry curvature. We identify four distinct MBPE mechanisms, including time-reversal-even and time-reversal-odd injection-like and shift-like photocurrents driven by linearly or circularly polarized light. Our symmetry analysis shows that the MBPE is allowed in broad classes of two-dimensional centrosymmetric nonmagnetic and magnetic materials under normally incident light. As representative examples, we predict sizable MBPEs in centrosymmetric monolayer Pt$_2$HgSe$_3$ and altermagnetic RuF$_4$, with current densities on the order of $1~\mathrm{mA/m}$. Notably, the injection-like photocurrent in RuF$_4$ reverses sign upon switching either the light helicity or the Néel vector. Our results extend the bulk photovoltaic effect to centrosymmetric solids and open new opportunities for electrical readout of altermagnetic order.

cond-mat.mtrl-sci↗
arXiv2026

Simultaneous Sharp-Interface Limits for a Diffuse-Interface Landau--de Gennes Droplet Model

Nematic droplet models based on diffuse interfaces involve competing microscopic transition layers whose interaction is lost if the small parameters are sent to zero successively. We study this interaction for the diffuse-interface Landau--de Gennes model of Shi et al., with phase-field thickness $\varepsilon$ and exterior $Q$-tensor relaxation length $κ_\varepsilon$, and show that their relative scale selects the effective sharp-interface theory. Within a regular-interface class, we establish the corresponding sharp-interface limit for every limiting ratio $r=\lim_{\varepsilon\to0}κ/\varepsilon\in[0,+\infty]$. Three qualitatively different regimes emerge. When $r=+\infty$, the two transition layers separate asymptotically and the interfacial costs decouple, recovering the previously obtained iterated limit. When $r\in(0,+\infty)$, the layers overlap and produce a genuinely coupled, $r$-dependent anisotropic surface density. When $r=0$, finite energy forces the interior trace of $Q$ to vanish on the limiting interface. The comparable-scale regime requires a genuinely coupled interfacial analysis. Thus the exterior relaxation length is not merely a regularization parameter: together with the phase-field thickness, it determines the macroscopic surface energy and anchoring law of the sharp-interface limit.

math.AP↗
arXiv2026

Modular Landau Analysis of Coupled Phase Transitions: Application to Ferroelectric Liquid Crystals

In systems with coupled order parameters, the relaxation of existing order can modify subsequent transitions, while cooperation and competition between emerging modes can produce complex phase sequences. We develop a modular Landau analysis that locally relaxes noncritical parent-phase variables to obtain effective one- and two-mode free energies. The resulting coefficients quantify relaxation-induced changes in transition characteristics and mode coupling, including changes in transition order for even-symmetry modes. The resulting modules are assembled successively along parent branches, and the equilibrium phase sequence is selected by comparing the fully relaxed free energies of the admissible branches. We apply the framework to ferroelectric liquid crystals, where orientational, polar, smectic, and tilt orders interact and give rise to multiple competing pathways, using a uniaxial model with longitudinal polarization. We fit the model to the measurements reported by Strachan et al. [Adv. Sci. 12(3), 2409754 (2025)]. A single thermodynamic parameter set reproduces the observed transition temperatures, transition types, and phase sequence, while capturing the main birefringence trend. Smectic layer spacing is reconstructed separately through a fitted observation map based on the calculated equilibrium tilt.

cond-mat.soft↗
arXiv2026

SlimFormer++: Token Mixing Where It Matters for Efficient 3D Medical Image Segmentation

To determine where computation is needed in volumetric segmentation networks, we analyze block-level redundancy and stage-wise token-mixer contribution. Centered Kernel Alignment reveals high input-output representational similarity within blocks of representative convolutional, hybrid, and Transformer-style segmentation networks. Similarity-guided pruning reduces their parameter counts by 39-75% while decreasing mean Dice by less than 0.4 percentage points. To examine token mixers within blocks, we introduce Anatomical Relation Distortion, which quantifies how bypassing a mixer distorts feature relations between labeled anatomical positions. Applied to an unpruned candidate architecture, Centered Kernel Alignment indicates high representational redundancy in intermediate encoder mixers, whereas Anatomical Relation Distortion shows that they contribute little and the bottleneck mixer most. These findings motivate SlimFormer++, which uses Selective Mixing blocks with windowed Mamba mixing in the first encoder stage and only residual channel MLPs in deeper stages. At the bottleneck, Augmented Mixing injects similarity-weighted sparse long-range context into the axial taps of a depthwise convolution at cost linear in voxel count. Evaluated against 12 baselines across eight public CT, MRI, and ultrasound datasets (1,809 cases), SlimFormer++ achieves the highest mean Dice (87.84%) and lowest mean HD95 (6.50 mm) with the fewest parameters (5.98 million). These results support stage-aware token mixing as a design strategy for improving the accuracy-efficiency trade-off in volumetric segmentation. The source code is available at https://github.com/WHHHHHY822/SlimFormerPlusPlus

cs.CV↗
arXiv2026

MESS: Multi-Exposure Sequence Synthesis for Generalizable Image Enhancement

Despite the significant progress in image enhancement under adverse illuminations, such as low-light image enhancement, exposure correction, and backlit image enhancement, existing models show limited generalizability to different scenarios. A major bottleneck lies in the lack of large-scale and diverse training data, as these tasks typically rely on manually captured, precisely aligned image pairs that are expensive to scale up. Although synthetic data have been explored as an alternative, existing synthesis methods are limited in modeling complex adverse illumination conditions and camera imaging pipelines. To address this issue, we propose a novel framework to synthesize realistic training pairs for generalizable image enhancement in the wild. Specifically, we first derive multi-exposure sequences (MES) from high-bit RAW images by rendering them under different exposure settings through an emulated ISP pipeline. Using these RAW-derived MES as supervision, we train a diffusion-based RGB-to-MES generator to synthesize plausible exposure sequences from a single 8-bit RGB image, extending the synthesis pipeline from limited RAW collections to large-scale in-the-wild RGB data. For each source image, we randomly sample one synthesized exposure variant as the degraded input and use the original high-quality RGB image as the target. With the proposed MES Synthesis (MESS) approach, we construct a dataset of 500K training pairs and train a lightweight network, which demonstrates significantly better generalization performance than existing models across various image enhancement tasks. Code and data are available at https://github.com/PolyU-VCLab/MESS.

cs.CV↗
arXiv2026

Isolated singularities of solutions to the Yamabe equation in higher dimensions II

This is the second of two papers on isolated singularities of positive solutions to the Yamabe equation on manifolds with nonflat background metrics. In the first paper, we proved that, in dimensions $3\le n\le29$, every isolated singularity is either removable or the solution is asymptotic to a radial model singular solution in the flat metric on the Euclidean space with origin as its singular point---a Fowler solution. In the present paper, we show that this dimension range is sharp. More precisely, for each $n\ge30$, we construct a smooth metric on $\mathbb{RP}^n$ and a positive solution of the corresponding Yamabe equation having an isolated singularity, near which the solution blows up along a sequence at a rate much faster than that of Fowler solutions. Moreover, the singular solution defines a complete conformal metric of constant positive scalar curvature and the metric agrees with the round metric to infinite order at the singular point.

math.AP↗
arXiv2026

UBASE: An AI Search Engine for Trillion-Scale Vector Data Management at ByteDance

Since 2016, UBASE has been the foundation of ByteDance's search infrastructure, scaling to more than 7,000 clusters and 300 PB of indexed data. Driven by the demands of AI workloads, UBASE has evolved from a text search engine into a unified AI search system supporting vector retrieval, lexical matching, and predicate filtering. Its largest deployment indexes nearly one trillion high-dimensional vectors. This scale exposes two central bottlenecks in AI-era retrieval: memory-intensive graph-index construction under sustained ingestion, and the prohibitive cost of keeping vector indexes entirely in memory. UBASE addresses these bottlenecks with two techniques. First, it introduces a quantization-aware vector kernel based on SymRaBitQ, a new symmetric quantization scheme with tight theoretical guarantees that allows index construction to run directly in the quantized space accurately and efficiently without retaining a copy of full-precision vectors. Second, it provides a hybrid storage engine that supports memory-resident, hybrid, and SSD-resident deployments, with fine-grained record-level caching to trade memory for latency under operational control. On large-scale benchmarks, UBASE improves throughput by up to 3x, reduces indexing memory by 80%, and lowers serving cost by up to 99.4% compared with prior systems, while supporting trillion-vector scale, write-heavy or latency-sensitive workloads in production.

cs.DB↗
arXiv2026

Product Costing in Data Pipelines

Large data-intensive enterprises spend millions of dollars monthly running data pipelines. Apportioning this cost back to individual products, known as product costing, is a prerequisite for data pricing and cost recovery. However, this process is fundamentally more challenging than physical manufacturing costing due to the unique characteristics of data pipelines, including the non-rivalrous nature of data that complicates the costing formulation, the prevalence of cyclic dependencies that distort the accuracy of results, and the massive scale of pipelines that leads to computational intractability. In this paper, for the first time, we formalize the problem of product costing in data pipelines as a manufacturing cost of data products (MCDP) problem to overcome the non-rivalrous nature of data. To solve MCDP, we propose CostApp, a hybrid algorithm that achieves both exactness and scalability, even for cyclic and massive-scale pipelines. CostApp performs a dimensionality reduction by isolating cyclic dependencies via a feedback arc set (FAS). It aggregates costs onto a compact set of key vertices using a scalable iterative approach, transforming the intractable computational problem into a reduced linear system that can be solved exactly and efficiently by a basic matrix-based method. Experiments on both real-world data and synthetic data show that CostApp scales to process a production pipeline with over 28 million entities within an hour while preserving exactness, supporting pipelines up to 10^2 larger than the direct matrix-based baseline.

cs.CE↗
arXiv2026

Finite Time Noncollapsing Kähler-Ricci Flow: Gaussian Type Heat Kernel, Gromov-Hausdorff Limits

In this paper, we study heat kernel estimates along the Kähler-Ricci flow with finite time noncollapsing singularities over a compact Kähler manifold. The starting point is the type-II scalar-curvature estimate, which gives the natural parabolic scale $r\lesssim T-t$ near the singular time. By combining the distance distortion and space-time cut-off estimates of Bamler--Zhang with entropy, mean-value inequality, and conjugate heat-kernel estimates, we obtain two-sided Gaussian bounds and gradient bounds that are uniform on this scale. Under the global type-I scalar-curvature estimate, the same estimates hold on every scale. We then use these estimates and the monotonicity of the $W_1$-distance between conjugate heat flows to construct a canonical singular-time pseudo-distance. The distances $d_t$ (induced by $ω(t)$) converge uniformly to this pseudo-distance. Its metric quotient is the unique Gromov--Hausdorff limit. Finally, we apply Bamler's theory and the recent work of Hallgren--Zhang to obtain the following. Every fixed-base-point type-I blow-up admits a subsequential metric-flow limit whose time slices are singular shrinking Kähler--Ricci soliton spaces with normal complex-analytic and log-terminal structure, unique tangent cones, and a polarized Fano fibration. For moving base points, the limiting metric flow, equipped with its limiting conjugate heat flow, is a metric soliton precisely under an entropy-pinching condition.

math.DG↗
arXiv2026

Pack Light, CarryOn: Building Object-Oriented Context for Long-Horizon Agents

Long-horizon agents accumulate progress in persistent task state, yet repeatedly process interaction histories describing how that progress was achieved. This raises a natural question: can long-horizon context instead be organized around reusable progress rather than the history that produced it? In this paper, we introduce CarryOn, an object-oriented context runtime that organizes continuation around reusable task objects. Rather than treating compressed histories or reconstructed accounts of progress as the primary basis for continuation, CarryOn builds a context view from continuation carriers, complementary execution context, and a live tail of recent interactions. A core challenge is keeping this reusable basis valid after its surrounding history is retired: how can CarryOn update the basis without reconstructing the full representation at every fold? CarryOn maintains carrier records across folds, revising affected records against bound evidence and applicable environment checks while carrying forward unaffected records and jointly updating execution context. Experiments on OSWorld 2.0 and BrowseComp-Plus show improved gains of 5.8% and 7.0% while reducing total token usage by 51.9% and 22.9%, respectively. Project page: https://ethannju.github.io/carryon

cs.AI↗
arXiv2026

EvoWeave: Composing Reusable Agent Experience through Operation Level Graphs

Different tasks often require interactive agents to perform similar operations at intermediate steps. Retrieving an entire past workflow can introduce irrelevant steps, whereas retrieving isolated operations can discard the dependencies needed to reuse them. We introduce EvoWeave, a framework for experience memory that represents abstract states as nodes and reusable operations as directed edges with input and output contracts, provenance, and capability references. A planner decomposes the current task into a dependency graph; operation retrieval and bounded directed search then assemble relevant experience fragments. An execution agent can inspect these fragments and their referenced capabilities, while reflection separates concrete execution evidence from reusable abstractions. We study the framework across four benchmarks, five evaluation settings, and three language model backbones. Across backbones, AppWorld Normal and Challenge task goal completion average 51.91 and 27.99, respectively, exceeding the baseline without memory by 6.99 and 4.99 percentage points. An AppWorld ablation associates directed connections with an additional improvement of 2.08 points over independent operation cards. The results motivate composition of reusable operations while exposing tradeoffs in retrieval overhead and conditions for valid reuse.

cs.AI↗
arXiv2026

Text-Vision Co-Instructed Image Editing

Existing image editing methods can be generally categorized into textual instruction-based and visual prompt-based ones. Textual instructions are semantically expressive, but are limited by the coarse granularity of spatial control of the editing results. In contrast, visual prompts such as drag and point can provide precise spatial guidance, but are limited by the inherent ambiguity in semantic intent. To unify the strength of textual and visual prompts, we present Text-Vision Co-Instructed Image Editing, which jointly models textual instructions as semantic intent and sparse visual instructions as spatial guidance, aiming to achieve precise and intent-faithful image manipulation. To this end, we first construct a textual-visual instruction paired dataset with more than 23K samples derived from dynamic videos, enabling aligned supervision for cross-modal instruction. We then propose TV-Edit, a Textual-Visual instruction unified Editing framework to contextualize drag or point-based visual instructions with image-text semantics and lift them into semantic-aware control representations for pretrained editing backbones. By integrating semantic intent and spatial constraints, TV-Edit leads to more precise spatial control, less instruction ambiguity, and stronger structural consistency than text-only or drag-based alternatives. Finally, we establish TV-Edit-Bench, a deliberately designed benchmark to evaluate semantic faithfulness, spatial alignment, and visual consistency with ground-truth references and controlled textual-visual variations for reliable assessment. Our experiments across multiple editing backbones demonstrate that TV-Edit consistently yields more precise and intent-faithful edits, significantly outperforming state-of-the-art instruction-based and drag-based baselines.

cs.CV↗
arXiv2026

HaloBlock: Faster and Better Decoding for Diffusion Large Language Models

Diffusion large language models (dLLMs) have attracted increasing research attention due to their bidirectional attention and arbitrary order decoding. However, there are two issues in dLLMs inference: the decoding difficulty is dominated by the tokens on the right side of the block, and the block-wise decoding inherently suffers from boundary contextual discontinuity. To tackle these limitations, we propose HaloBlock, a difficulty-aware adaptive halo mechanism for dLLM inference. We first leverage bidirectional cumulative entropy to characterize block decoding difficulty. Then, based on the bidirectional decoding difficulty, we adaptively allocate bidirectional halo regions on both sides of each block to mitigate boundary contextual discontinuity. The difficulty-aware design of HaloBlock smooths cross-block context propagation and lowers decoding difficulty for hard tokens, yielding better generation quality and faster convergence. Evaluated on LLaDA-8B-Instruct, Dream-7B-Instruct and LLaDA-1.5 across mathematical reasoning and code generation benchmarks, HaloBlock demonstrates consistent accuracy improvements and higher inference efficiency. Source codes are publicly available at https://github.com/PolyU-VCLab/HaloBlock.

cs.LG↗
arXiv2026

Tanchun: Robust Flow-Based HTTP(S) DDoS Detection via Behavioral Attribute Guidance

HTTP(S) DDoS attacks have surged in frequency and scale, posing a significant threat to network security. Flow-based detection is a promising solution for large-scale HTTP(S) DDoS detection due to its decryption-free operation and scalability. However, the coarse-grained numerical statistics in flow data limit accurate attack detection and lack robustness when benign and attack traffic patterns drift over time. In this paper, we present Tanchun, a drift-aware flow-based HTTP(S) DDoS detection system that relies solely on decryption-free flow data. Specifically, Tanchun groups flows by IP-pair and aggregates them over consecutive time slots into behavior sequences, capturing temporal access patterns to compensate for the limited information in individual flows. It further abstracts these sequences into human-understandable behavioral attributes that capture stable behavioral semantics. Tanchun then aligns behavior sequences with their corresponding behavioral attributes to guide representation learning, enabling robust detection under traffic drift. We evaluate Tanchun across diverse HTTP(S) DDoS scenarios. Tanchun achieves the best overall performance, with an average F1-score of 0.988, and outperforms the second-best method by 5.3\% and 9.6\% under benign-traffic and attack-traffic drift, respectively. Experiments on real-world traffic traces further demonstrate its practical effectiveness.

cs.NI↗