Search arXiv⌕ Search

arXiv subjects

Chao Zhang

Publications and source records attributed to Chao Zhang.

At least 217 records · Page 12Linked to original sources
arXiv2026

Precise Attribute Intensity Control in Large Language Models via Targeted Representation Editing

Precise attribute intensity control--generating Large Language Model (LLM) outputs with specific, user-defined attribute intensities--is crucial for AI systems adaptable to diverse user expectations. Current LLM alignment methods, however, typically provide only directional or open-ended guidance, failing to reliably achieve exact attribute intensities. We address this limitation with three key designs: (1) reformulating precise attribute intensity control as a target-reaching problem, rather than simple maximization; (2) training a lightweight value function via temporal-difference learning to predict final attribute intensity scores from partial generations, thereby steering LLM outputs; and (3) employing gradient-based interventions on hidden representations to navigate the model precisely towards specific attribute intensity targets. Our method enables fine-grained, continuous control over attribute intensities, moving beyond simple directional alignment. Experiments on LLaMA-3.2-3b and Phi-4-mini confirm our method's ability to steer text generation to user-specified attribute intensities with high accuracy. Finally, we demonstrate efficiency enhancements across three downstream tasks: preference data synthesis, Pareto frontier approximation and optimization, and distillation of aligned behaviors for intervention-free inference. Our code is available on https://github.com/Pre-Control/pre-control

cs.AI↗
arXiv2026

SecCodeBench-V2 Technical Report

We introduce SecCodeBench-V2, a publicly released benchmark for evaluating Large Language Model (LLM) copilots' capabilities of generating secure code. SecCodeBench-V2 comprises 98 generation and fix scenarios derived from Alibaba Group's industrial productions, where the underlying security issues span 22 common CWE (Common Weakness Enumeration) categories across five programming languages: Java, C, Python, Go, and JavaScript. SecCodeBench-V2 adopts a function-level task formulation: each scenario provides a complete project scaffold and requires the model to implement or patch a designated target function under fixed interfaces and dependencies. For each scenario, SecCodeBench-V2 provides executable proof-of-concept (PoC) test cases for both functional validation and security verification. All test cases are authored and double-reviewed by security experts, ensuring high fidelity, broad coverage, and reliable ground truth. Beyond the benchmark itself, we build a unified evaluation pipeline that assesses models primarily via dynamic execution. For most scenarios, we compile and run model-generated artifacts in isolated environments and execute PoC test cases to validate both functional correctness and security properties. For scenarios where security issues cannot be adjudicated with deterministic test cases, we additionally employ an LLM-as-a-judge oracle. To summarize performance across heterogeneous scenarios and difficulty levels, we design a Pass@K-based scoring protocol with principled aggregation over scenarios and severity, enabling holistic and comparable evaluation across models. Overall, SecCodeBench-V2 provides a rigorous and reproducible foundation for assessing the security posture of AI coding assistants, with results and artifacts released at https://alibaba.github.io/sec-code-bench. The benchmark is publicly available at https://github.com/alibaba/sec-code-bench.

cs.CR↗
arXiv2026

sleep2vec: Unified Cross-Modal Alignment for Heterogeneous Nocturnal Biosignals

Tasks ranging from sleep staging to clinical diagnosis traditionally rely on standard polysomnography (PSG) devices, bedside monitors and wearable devices, which capture diverse nocturnal biosignals (e.g., EEG, EOG, ECG, SpO$_2$). However, heterogeneity across devices and frequent sensor dropout pose significant challenges for unified modelling of these multimodal signals. We present \texttt{sleep2vec}, a foundation model for diverse and incomplete nocturnal biosignals that learns a shared representation via cross-modal alignment. \texttt{sleep2vec} is contrastively pre-trained on 42,249 overnight recordings spanning nine modalities using a \textit{Demography, Age, Site \& History-aware InfoNCE} objective that incorporates physiological and acquisition metadata (\textit{e.g.}, age, gender, recording site) to dynamically weight negatives and mitigate cohort-specific shortcuts. On downstream sleep staging and clinical outcome assessment, \texttt{sleep2vec} consistently outperforms strong baselines and remains robust to any subset of available modalities and sensor dropout. We further characterize, to our knowledge for the first time, scaling laws for nocturnal biosignals with respect to modality diversity and model capacity. Together, these results show that unified cross-modal alignment, coupled with principled scaling, enables label-efficient, general-purpose modelling of real-world nocturnal biosignals.

cs.LG↗
arXiv2026

DMP-3DAD: Cross-Category 3D Anomaly Detection via Realistic Depth Map Projection with Few Normal Samples

Cross-category anomaly detection for 3D point clouds aims to determine whether an unseen object belongs to a target category using only a few normal examples. Most existing methods rely on category-specific training, which limits their flexibility in few-shot scenarios. In this paper, we propose DMP-3DAD, a training-free framework for cross-category 3D anomaly detection based on multi-view realistic depth map projection. Specifically, by converting point clouds into a fixed set of realistic depth images, our method leverages a frozen CLIP visual encoder to extract multi-view representations and performs anomaly detection via weighted feature similarity, which does not require any fine-tuning or category-dependent adaptation. Extensive experiments on the ShapeNetPart dataset demonstrate that DMP-3DAD achieves state-of-the-art performance under few-shot setting. The results show that the proposed approach provides a simple yet effective solution for practical cross-category 3D anomaly detection.

cs.CV↗
arXiv2026

Speech-Audio Compositional Attacks on Multimodal LLMs and Their Mitigation with SALMONN-Guard

Recent progress in LLMs has enabled understanding of audio signals, but has also exposed new safety risks arising from complex audio inputs that are inadequately handled by current safeguards. We introduce SACRED-Bench (Speech-Audio Composition for RED-teaming) to evaluate the robustness of LLMs under complex audio-based attacks. Unlike existing perturbation-based methods that rely on noise optimization or white-box access, SACRED-Bench exploits speech-audio composition to enable effective black-box attacks. SACRED-Bench adopts three composition mechanisms: (a) overlap of harmful and benign speech, (b) mixture of benign speech with harmful non-speech audio, and (c) multi-speaker dialogue. These mechanisms focus on evaluating safety in settings where benign and harmful intents co-occur within a single auditory scene. Moreover, questions in SACRED-Bench are designed to implicitly refer to content in the audio, such that no explicit harmful information appears in the text prompt alone. Experiments demonstrate that even Gemini 2.5 Pro, a state-of-the-art proprietary LLM with safety guardrails fully enabled, still exhibits a 66% attack success rate. To bridge this gap, we propose SALMONN-Guard, the first guard model that jointly inspects speech, audio, and text for safety judgments, reducing the attack success rate to 20%. Our results highlight the need for audio-aware defenses to ensure the safety of multimodal LLMs. The dataset and SALMONN-Guard checkpoints can be found at https://huggingface.co/datasets/tsinghua-ee/SACRED-Bench.

cs.SD↗
arXiv2026

Conditional Sequence Modeling for Safe Reinforcement Learning

Offline safe reinforcement learning (RL) aims to learn policies from a fixed dataset while maximizing performance under cumulative cost constraints. In practice, deployment requirements often vary across scenarios, necessitating a single policy that can adapt zero-shot to different cost thresholds. However, most existing offline safe RL methods are trained under a pre-specified threshold, yielding policies with limited generalization and deployment flexibility across cost thresholds. Motivated by recent progress in conditional sequence modeling (CSM), which enables flexible goal-conditioned control by specifying target returns, we propose RCDT, a CSM-based method that supports zero-shot deployment across multiple cost thresholds within a single trained policy. RCDT is the first CSM-based offline safe RL algorithm that integrates a Lagrangian-style cost penalty with an auto-adaptive penalty coefficient. To avoid overly conservative behavior and achieve a more favorable return--cost trade-off, a reward--cost-aware trajectory reweighting mechanism and Q-value regularization are further incorporated. Extensive experiments on the DSRL benchmark demonstrate that RCDT consistently improves return--cost trade-offs over representative baselines, advancing the state-of-the-art in offline safe RL.

cs.LG↗
arXiv2026

D-ORCA: Dialogue-Centric Optimization for Robust Audio-Visual Captioning

Spoken dialogue is a primary source of information in videos; therefore, accurately identifying who spoke what and when is essential for deep video understanding. We introduce D-ORCA, a \textbf{d}ialogue-centric \textbf{o}mni-modal large language model optimized for \textbf{r}obust audio-visual \textbf{ca}ptioning. We further curate DVD, a large-scale, high-quality bilingual dataset comprising nearly 40,000 multi-party dialogue videos for training and 2000 videos for evaluation in English and Mandarin, addressing a critical gap in the open-source ecosystem. To ensure fine-grained captioning accuracy, we adopt group relative policy optimization with three novel reward functions that assess speaker attribution accuracy, global speech content accuracy, and sentence-level temporal boundary alignment. These rewards are derived from evaluation metrics widely used in speech processing and, to our knowledge, are applied for the first time as reinforcement learning objectives for audio-visual captioning. Extensive experiments demonstrate that D-ORCA substantially outperforms existing open-source models in speaker identification, speech recognition, and temporal grounding. Notably, despite having only 8 billion parameters, D-ORCA achieves performance competitive with Qwen3-Omni across several general-purpose audio-visual understanding benchmarks. Demos are available at \href{https://d-orca-llm.github.io/}{https://d-orca-llm.github.io/}. Our code, data, and checkpoints will be available at \href{https://github.com/WeChatCV/D-ORCA/}{https://github.com/WeChatCV/D-ORCA/}.

cs.CV↗
arXiv2026

Probing Quantum Gravity effects with Extreme Mass Ratio Inspirals around Rotating Hayward Black Holes

We investigate extreme mass-ratio inspirals (EMRIs) around a rotating Hayward black hole to assess the detectability of signatures arising from quantum gravity.The quantum parameter $α_0$, which encodes deviations from general relativity (GR), introduces extra correction terms in both the orbital frequency and the fluxes. Our results show that after one year of accumulated observation, these corrections induce a detectable dephasing in the EMRI waveform. Using the modified orbital evolution driven by $α_0$, we generate waveforms via the augmented analytic kludge (AAK) model implemented in the \texttt{FastEMRIWaveforms} package. Furthermore, we utilize the time-delay interferometry (TDI) to suppress the laser noise and phase fluctuations induced by spacecraft motion, and then employ the Fisher information matrix (FIM) to test the sensitivity of LISA in detecting deviations from GR. Our results demonstrate the potential of LISA to probe quantum-gravity effects through high-precision observations of EMRIs.

gr-qc↗
arXiv2026

Mobile impurity coupled to correlated lattice bosons

We investigate how the coherence and spatial dressing of a single impurity evolve in the two-dimensional Bose-Hubbard model when the impurity couples attractively to the bath. Using large-scale, sign-problem-free worm-algorithm quantum Monte Carlo, we measure the impurity winding, bath superfluid response and compressibility, and impurity-bath density correlations. In a compressible superfluid bath ($U_{\mathrm{b}}/t=13.3$), strengthening attraction drives an interaction-controlled \emph{winding-collapse self-trapping crossover}: a mobile light polaron evolves continuously into a heavy polaron and ultimately into a bound cluster with vanishing winding, while the bath remains globally superfluid. In incompressible Mott-insulating baths, by contrast, extended density rearrangements are suppressed and the dressing cloud collapses; we compare the resulting short-range deformation patterns for both attractive and repulsive couplings. Across the SF-MI transition at fixed moderate attraction ($U_{\mathrm{ib}}/t=-8.0$), the impurity crosses over from a mobile polaron with an extended deformation cloud to a nearly free defect with minimal dressing in the Mott background; for this coupling it does not lock into a fully self-trapped defect even deep in the insulator. Together with our companion Letter [Impurity Self-Trapping in Lattice Bose systems] on repulsive couplings, these results provide a unified microscopic picture of impurity self-trapping in correlated lattice bosons, connecting interaction-driven winding collapse in the superfluid to compressibility-controlled undressing across the SF-MI transition.

cond-mat.quant-gas↗
arXiv2026

HoliAntiSpoof: Audio LLM for Holistic Speech Anti-Spoofing

Recent advances in speech synthesis and editing have made speech spoofing increasingly challenging. However, most existing methods treat spoofing as binary classification, overlooking that diverse spoofing techniques manipulate multiple, coupled speech attributes and their semantic effects. In this paper, we introduce HoliAntiSpoof, the first audio large language model (ALLM) framework for holistic speech anti-spoofing analysis. HoliAntiSpoof reformulates spoofing analysis as a unified text generation task, enabling joint reasoning over spoofing methods, affected speech attributes, and their semantic impacts. To support semantic-level analysis, we introduce DailyTalkEdit, a new anti-spoofing benchmark that simulates realistic conversational manipulations and provides annotations of semantic influence. Extensive experiments demonstrate that HoliAntiSpoof outperforms conventional baselines across multiple settings, while preliminary results show that in-context learning further improves out-of-domain generalization. These findings indicate that ALLMs not only enhance speech spoofing detection performance but also enable interpretable analysis of spoofing behaviors and their semantic effects, pointing towards more trustworthy and explainable speech security. Data and code are publicly available.

cs.SD↗
arXiv2026

ERNIE 5.0 Technical Report

In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are trained from scratch under a unified next-group-of-tokens prediction objective, based on an ultra-sparse mixture-of-experts (MoE) architecture with modality-agnostic expert routing. To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm. Within a single pre-training run, the model learns a family of sub-models with varying depths, expert capacities, and routing sparsity, enabling flexible trade-offs among performance, model size, and inference latency in memory- or time-constrained scenarios. Moreover, we systematically address the challenges of scaling reinforcement learning to unified foundation models, thereby guaranteeing efficient and stable post-training under ultra-sparse MoE architectures and diverse multimodal settings. Extensive experiments demonstrate that ERNIE 5.0 achieves strong and balanced performance across multiple modalities. To the best of our knowledge, among publicly disclosed models, ERNIE 5.0 represents the first production-scale realization of a trillion-parameter unified autoregressive model that supports both multimodal understanding and generation. To facilitate further research, we present detailed visualizations of modality-agnostic expert routing in the unified model, alongside comprehensive empirical analysis of elastic training, aiming to offer profound insights to the community.

cs.CL↗
arXiv2026

video-SALMONN S: Memory-Enhanced Streaming Audio-Visual LLM

Long-duration streaming video understanding is fundamental for future AI agents, yet remains limited by ineffective long-term memory. We introduce video-SALMONN S, a memory-enhanced streaming audio-visual large language model that processes over 3-hour videos at 1 FPS and 360p resolution, outperforming strong non-streaming models under the same memory budget. In addition to token merging or downsampling, video-SALMONN S is the first to employ test-time training (TTT) as a streaming memory mechanism for video understanding. TTT continuously transforms short-term multimodal representations into long-term memory embedded in model parameters. To improve long-range dependency modeling and memory capacity, we propose (i) a TTT_MEM layer with an additional long-span prediction objective, (ii) a two-stage training scheme, and (iii) a modality-aware memory reader. We further introduce the Episodic Learning from Video Memory (ELViM) benchmark, simulating agent-like scenarios where models must learn from videos observed hours earlier. video-SALMONN S consistently outperforms both streaming and non-streaming baselines by 3-7% on long video benchmarks. Notably, video-SALMONN S achieves a 15% absolute accuracy improvement over strong non-streaming models on ELViM, demonstrating strong learning abilities from video memory.

cs.CV↗
arXiv2026

LongCat-Flash-Thinking-2601 Technical Report

We introduce LongCat-Flash-Thinking-2601, a 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model with superior agentic reasoning capability. LongCat-Flash-Thinking-2601 achieves state-of-the-art performance among open-source models on a wide range of agentic benchmarks, including agentic search, agentic tool use, and tool-integrated reasoning. Beyond benchmark performance, the model demonstrates strong generalization to complex tool interactions and robust behavior under noisy real-world environments. Its advanced capability stems from a unified training framework that combines domain-parallel expert training with subsequent fusion, together with an end-to-end co-design of data construction, environments, algorithms, and infrastructure spanning from pre-training to post-training. In particular, the model's strong generalization capability in complex tool-use are driven by our in-depth exploration of environment scaling and principled task construction. To optimize long-tailed, skewed generation and multi-turn agentic interactions, and to enable stable training across over 10,000 environments spanning more than 20 domains, we systematically extend our asynchronous reinforcement learning framework, DORA, for stable and efficient large-scale multi-environment training. Furthermore, recognizing that real-world tasks are inherently noisy, we conduct a systematic analysis and decomposition of real-world noise patterns, and design targeted training procedures to explicitly incorporate such imperfections into the training process, resulting in improved robustness for real-world applications. To further enhance performance on complex reasoning tasks, we introduce a Heavy Thinking mode that enables effective test-time scaling by jointly expanding reasoning depth and width through intensive parallel thinking.

cs.AI↗
arXiv2026

Adaptive Dual-Weighting Framework for Federated Learning via Out-of-Distribution Detection

Federated Learning (FL) enables collaborative model training across large-scale distributed service nodes while preserving data privacy, making it a cornerstone of intelligent service systems in edge-cloud environments. However, in real-world service-oriented deployments, data generated by heterogeneous users, devices, and application scenarios are inherently non-IID. This severe data heterogeneity critically undermines the convergence stability, generalization ability, and ultimately the quality of service delivered by the global model. To address this challenge, we propose FLood, a novel FL framework inspired by out-of-distribution (OOD) detection. FLood dynamically counteracts the adverse effects of heterogeneity through a dual-weighting mechanism that jointly governs local training and global aggregation. At the client level, it adaptively reweights the supervised loss by upweighting pseudo-OOD samples, thereby encouraging more robust learning from distributionally misaligned or challenging data. At the server level, it refines model aggregation by weighting client contributions according to their OOD confidence scores, prioritizing updates from clients with higher in-distribution consistency and enhancing the global model's robustness and convergence stability. Extensive experiments across multiple benchmarks under diverse non-IID settings demonstrate that FLood consistently outperforms state-of-the-art FL methods in both accuracy and generalization. Furthermore, FLood functions as an orthogonal plug-in module: it seamlessly integrates with existing FL algorithms to boost their performance under heterogeneity without modifying their core optimization logic. These properties make FLood a practical and scalable solution for deploying reliable intelligent services in real-world federated environments.

cs.LG↗
arXiv2026

Stochastic smoothing accelerated gradient method for general constrained nonsmooth convex composite optimization

We propose a novel stochastic smoothing accelerated gradient (SSAG) method for general constrained nonsmooth convex composite optimization, and analyze the convergence rates. The SSAG method allows various smoothing techniques, and can deal with the nonsmooth term that is not easy to compute its proximal term, or that does not own the linear max structure. To the best of our knowledge, it is the first time to develop a stochastic approximation type method that treats the maximization of finite but numerous nonsmooth convex functions as a stochastic function, which significantly improves the computational efficiency. We prove that the SSAG method can simultaneously achieve the best-known order ${\cal{O}}(\frac{1}ε)$ of iteration complexity, and the optimal order ${\cal{O}}(\frac{1}{ε^2})$ of $\cal{SFO}$ complexity, using variable sample-size. Numerical results on the application arising from the distributionally robust optimization demonstrate the effectiveness and efficiency of the proposed SSAG method.

math.OC↗
arXiv2026

Probing Star-Forming Properties via ALMA Observations of Massive Protocluster IRAS 15596-5301

To deepen our understanding of star-forming properties, we studied a massive protocluster IRAS 15596-5301 using ALMA 870 um and 3 mm data. High-resolution 870 um data reveal 34 dense cores, including 3 hot molecular cores, with subsequent line surveys detecting 22 molecular species toward them. Two velocity components (I15596-red/I15596-blue) were found in the averaged H13CO+(1-0) spectrum, and two filaments were identified from velocity-resolved integrated intensity maps. A spatial overlap between the two filaments was observed, and this overlapping region exhibits a distinct bridge-shaped feature in the position-velocity diagram constructed along the entire filamentary structures. Combined with the reduced H13CO+/HCO+ ratio in the overlapping region and the three-dimensional position-position-velocity cube data, we conclude that a non-head-on collision occurs between the edges of the two filamentary structures in IRAS 15596-5301. Cluster analysis demonstrates that clusters located in the collision region host more evolved chemical rich dense cores than their counterparts in other regions. Our results thus indicate that star formation in I15596 is triggered or accelerated by a mild non-head-on collision between two filaments.

astro-ph.GA↗
arXiv2026

Investigating the transverse-momentum- and pseudorapidity-dependent flow vector decorrelation in p--Pb collisions with a Multi-Phase Transport model

The event-by-event fluctuations in the initial energy density of the nuclear collisions lead to the decorrelation of second order flow vector, as known as its transverse-momentum ($p_{\mathrm{T}}$) and pseudorapidity ($η$) dependence as observed in high-energy heavy-ion collisions. Existing measurements at the CERN Large Hadron Collider shown that these decorrelations are also observed in small collision systems. In this work, a systematic study of the transverse-momentum- and pseudorapidity-dependent flow vector decorrelation is performed in p--Pb collisions at the 5.02 TeV with A Multi-Phase Transport (AMPT) model using different tunings of the initial conditions, partonic and hadronic interactions. It is found that the string-melting version of the AMPT model provides a reasonable description of the measured flow vector decorrelation as a function of $p_{\mathrm{T}}$ and $η$. We demonstrate that the hadronic scatterings do not have significant impact on decorrelation in p--Pb collisions for different centrality selections, while both initial conditions and partonic interactions influence the magnitude of the decorrelations. In addition, we found that the subtraction of the nonflow, especially the long-range jet correlation, is crucial for the accurate extraction of the flow vector decorrelation in small collision systems. The comparison of data and model presented in this paper provide further insights in understanding the fluctuations of the flow vector with $p_{\mathrm{T}}$ and $η$ in small collision systems and has referential value for future measurements.

hep-ph↗
arXiv2026

OCR-Enhanced Multimodal ASR Can Read While Listening

Visual information, such as subtitles in a movie, often helps automatic speech recognition. In this paper, we propose Donut-Whisper, an audio-visual ASR model with dual encoder to leverage visual information to improve speech recognition performance in both English and Chinese. Donut-Whisper combines the advantage of the linear and the Q-Former-based modality alignment structures via a cross-attention module, generating more powerful audio-visual features. Meanwhile, we propose a lightweight knowledge distillation scheme showcasing the potential of using audio-visual models to teach audio-only models to achieve better performance. Moreover, we propose a new multilingual audio-visual speech recognition dataset based on movie clips containing both Chinese and English partitions. As a result, Donut-Whisper achieved significantly better performance on both English and Chinese partition of the dataset compared to both Donut and Whisper large V3 baselines. In particular, an absolute 5.75% WER reduction and a 16.5% absolute CER reduction were achieved on the English and Chinese sets respectively compared to the Whisper ASR baseline.

cs.SD↗