arXiv subjects

Dacheng Tao

Publications and source records attributed to Dacheng Tao.

At least 199 records · Page 11Linked to original sources

arXiv2025

MAPO: Mixed Advantage Policy Optimization

Recent advances in reinforcement learning for foundation models, such as Group Relative Policy Optimization (GRPO), have significantly improved the performance of foundation models on reasoning tasks. Notably, the advantage function serves as a central mechanism in GRPO for ranking the trajectory importance. However, existing explorations encounter both advantage reversion and advantage mirror problems, which hinder the reasonable advantage allocation across different query samples. In this work, we propose an easy but effective GRPO strategy, Mixed Advantage Policy Optimization (MAPO). We reveal that the trajectory appears with different certainty and propose the advantage percent deviation for samples with high-certainty trajectories. Furthermore, we dynamically reweight the advantage function for samples with varying trajectory certainty, thereby adaptively configuring the advantage function to account for sample-specific characteristics. Comparison with related state-of-the-art methods, along with ablation studies on different advantage variants, validates the effectiveness of our approach.

cs.AI↗

arXiv2025

Responsible Diffusion: A Comprehensive Survey on Safety, Ethics, and Trust in Diffusion Models

Diffusion models (DMs) have been investigated in various domains due to their ability to generate high-quality data, thereby attracting significant attention. However, similar to traditional deep learning systems, there also exist potential threats to DMs. To provide advanced and comprehensive insights into safety, ethics, and trust in DMs, this survey comprehensively elucidates its framework, threats, and countermeasures. Each threat and its countermeasures are systematically examined and categorized to facilitate thorough analysis. Furthermore, we introduce specific examples of how DMs are used, what dangers they might bring, and ways to protect against these dangers. Finally, we discuss key lessons learned, highlight open challenges related to DM security, and outline prospective research directions in this critical field. This work aims to accelerate progress not only in the technical capabilities of generative artificial intelligence but also in the maturity and wisdom of its application.

cs.CR↗

arXiv2025

EchoBench: Benchmarking Sycophancy in Medical Large Vision-Language Models

Recent benchmarks for medical Large Vision-Language Models (LVLMs) emphasize leaderboard accuracy, overlooking reliability and safety. We study sycophancy -- models' tendency to uncritically echo user-provided information -- in high-stakes clinical settings. We introduce EchoBench, a benchmark to systematically evaluate sycophancy in medical LVLMs. It contains 2,122 images across 18 departments and 20 modalities with 90 prompts that simulate biased inputs from patients, medical students, and physicians. We evaluate medical-specific, open-source, and proprietary LVLMs. All exhibit substantial sycophancy; the best proprietary model (Claude 3.7 Sonnet) still shows 45.98% sycophancy, and GPT-4.1 reaches 59.15%. Many medical-specific models exceed 95% sycophancy despite only moderate accuracy. Fine-grained analyses by bias type, department, perceptual granularity, and modality identify factors that increase susceptibility. We further show that higher data quality/diversity and stronger domain knowledge reduce sycophancy without harming unbiased accuracy. EchoBench also serves as a testbed for mitigation: simple prompt-level interventions (negative prompting, one-shot, few-shot) produce consistent reductions and motivate training- and decoding-time strategies. Our findings highlight the need for robust evaluation beyond accuracy and provide actionable guidance toward safer, more trustworthy medical LVLMs.

cs.CV↗

arXiv2025

Detoxifying Large Language Models via Autoregressive Reward Guided Representation Editing

Large Language Models (LLMs) have demonstrated impressive performance across various tasks, yet they remain vulnerable to generating toxic content, necessitating detoxification strategies to ensure safe and responsible deployment. Test-time detoxification methods, which typically introduce static or dynamic interventions into LLM representations, offer a promising solution due to their flexibility and minimal invasiveness. However, current approaches often suffer from imprecise interventions, primarily due to their insufficient exploration of the transition space between toxic and non-toxic outputs. To address this challenge, we propose \textsc{A}utoregressive \textsc{R}eward \textsc{G}uided \textsc{R}epresentation \textsc{E}diting (ARGRE), a novel test-time detoxification framework that explicitly models toxicity transitions within the latent representation space, enabling stable and precise reward-guided editing. ARGRE identifies non-toxic semantic directions and interpolates between toxic and non-toxic representations to reveal fine-grained transition trajectories. These trajectories transform sparse toxicity annotations into dense training signals, enabling the construction of an autoregressive reward model that delivers stable and precise editing guidance. At inference, the reward model guides an adaptive two-step editing process to obtain detoxified representations: it first performs directional steering based on expected reward gaps to shift representations toward non-toxic regions, followed by lightweight gradient-based refinements. Extensive experiments across 8 widely used LLMs show that ARGRE significantly outperforms leading baselines in effectiveness (-62.21% toxicity) and efficiency (-47.58% inference time), while preserving the core capabilities of the original model with minimal degradation. Our code is available at the website.

cs.CL↗

arXiv2025

Runaway is Ashamed, But Helpful: On the Early-Exit Behavior of Large Language Model-based Agents in Embodied Environments

Agents powered by large language models (LLMs) have demonstrated strong planning and decision-making capabilities in complex embodied environments. However, such agents often suffer from inefficiencies in multi-turn interactions, frequently trapped in repetitive loops or issuing ineffective commands, leading to redundant computational overhead. Instead of relying solely on learning from trajectories, we take a first step toward exploring the early-exit behavior for LLM-based agents. We propose two complementary approaches: 1. an $\textbf{intrinsic}$ method that injects exit instructions during generation, and 2. an $\textbf{extrinsic}$ method that verifies task completion to determine when to halt an agent's trial. To evaluate early-exit mechanisms, we introduce two metrics: one measures the reduction of $\textbf{redundant steps}$ as a positive effect, and the other evaluates $\textbf{progress degradation}$ as a negative effect. Experiments with 4 different LLMs across 5 embodied environments show significant efficiency improvements, with only minor drops in agent performance. We also validate a practical strategy where a stronger agent assists after an early-exit agent, achieving better performance with the same total steps. We will release our code to support further research.

cs.CL↗

arXiv2025

Efficient Learning for Linear Properties of Bounded-Gate Quantum Circuits

The vast and complicated large-qubit state space forbids us to comprehensively capture the dynamics of modern quantum computers via classical simulations or quantum tomography. Recent progress in quantum learning theory prompts a crucial question: can linear properties of a large-qubit circuit with d tunable RZ gates and G-d Clifford gates be efficiently learned from measurement data generated by varying classical inputs? In this work, we prove that the sample complexity scaling linearly in $d$ is required to achieve a small prediction error, while the corresponding computational complexity may scale exponentially in d. To address this challenge, we propose a kernel-based method leveraging classical shadows and truncated trigonometric expansions, enabling a controllable trade-off between prediction accuracy and computational overhead. Our results advance two crucial realms in quantum computation: the exploration of quantum algorithms with practical utilities and learning-based quantum system certification. We conduct numerical simulations to validate our proposals across diverse scenarios, encompassing quantum information processing protocols, Hamiltonian simulation, and variational quantum algorithms up to 60 qubits.

quant-ph↗

arXiv2025

One Flight Over the Gap: A Survey from Perspective to Panoramic Vision

Driven by the demand for spatial intelligence and holistic scene perception, omnidirectional images (ODIs), which provide a complete 360\textdegree{} field of view, are receiving growing attention across diverse applications such as virtual reality, autonomous driving, and embodied robotics. Despite their unique characteristics, ODIs exhibit remarkable differences from perspective images in geometric projection, spatial distribution, and boundary continuity, making it challenging for direct domain adaption from perspective methods. This survey reviews recent panoramic vision techniques with a particular emphasis on the perspective-to-panorama adaptation. We first revisit the panoramic imaging pipeline and projection methods to build the prior knowledge required for analyzing the structural disparities. Then, we summarize three challenges of domain adaptation: severe geometric distortions near the poles, non-uniform sampling in Equirectangular Projection (ERP), and periodic boundary continuity. Building on this, we cover 20+ representative tasks drawn from more than 300 research papers in two dimensions. On one hand, we present a cross-method analysis of representative strategies for addressing panoramic specific challenges across different tasks. On the other hand, we conduct a cross-task comparison and classify panoramic vision into four major categories: visual quality enhancement and assessment, visual understanding, multimodal understanding, and visual generation. In addition, we discuss open challenges and future directions in data, models, and applications that will drive the advancement of panoramic vision research. We hope that our work can provide new insight and forward looking perspectives to advance the development of panoramic vision technologies. Our project page is https://insta360-research-team.github.io/Survey-of-Panorama

cs.CV↗

arXiv2025

CoVeR: Conformal Calibration for Versatile and Reliable Autoregressive Next-Token Prediction

Autoregressive pre-trained models combined with decoding methods have achieved impressive performance on complex reasoning tasks. While mainstream decoding strategies such as beam search can generate plausible candidate sets, they often lack provable coverage guarantees, and struggle to effectively balance search efficiency with the need for versatile trajectories, particularly those involving long-tail sequences that are essential in certain real-world applications. To address these limitations, we propose \textsc{CoVeR}, a novel model-free decoding strategy wihtin the conformal prediction framework that simultaneously maintains a compact search space and ensures high coverage probability over desirable trajectories. Theoretically, we establish a PAC-style generalization bound, guaranteeing that \textsc{CoVeR} asymptotically achieves a coverage rate of at least $1 - α$ for any target level $α\in (0,1)$.

cs.LG↗

arXiv2025

Artificial intelligence for representing and characterizing quantum systems

Efficient characterization of large-scale quantum systems, especially those produced by quantum analog simulators and megaquop quantum computers, poses a central challenge in quantum science due to the exponential scaling of the Hilbert space with respect to system size. Recent advances in artificial intelligence (AI), with its aptitude for high-dimensional pattern recognition and function approximation, have emerged as a powerful tool to address this challenge. A growing body of research has leveraged AI to represent and characterize scalable quantum systems, spanning from theoretical foundations to experimental realizations. Depending on how prior knowledge and learning architectures are incorporated, the integration of AI into quantum system characterization can be categorized into three synergistic paradigms: machine learning, and, in particular, deep learning and language models. This review discusses how each of these AI paradigms contributes to two core tasks in quantum systems characterization: quantum property prediction and the construction of surrogates for quantum states. These tasks underlie diverse applications, from quantum certification and benchmarking to the enhancement of quantum algorithms and the understanding of strongly correlated phases of matter. Key challenges and open questions are also discussed, together with future prospects at the interface of AI and quantum science.

quant-ph↗

arXiv2025

Rethinking Data Protection in the (Generative) Artificial Intelligence Era

The (generative) artificial intelligence (AI) era has profoundly reshaped the meaning and value of data. No longer confined to static content, data now permeates every stage of the AI lifecycle from the training samples that shape model parameters to the prompts and outputs that drive real-world model deployment. This shift renders traditional notions of data protection insufficient, while the boundaries of what needs safeguarding remain poorly defined. Failing to safeguard data in AI systems can inflict societal and individual, underscoring the urgent need to clearly delineate the scope of and rigorously enforce data protection. In this perspective, we propose a four-level taxonomy, including non-usability, privacy preservation, traceability, and deletability, that captures the diverse protection needs arising in modern (generative) AI models and systems. Our framework offers a structured understanding of the trade-offs between data utility and control, spanning the entire AI pipeline, including training datasets, model weights, system prompts, and AI-generated content. We analyze representative technical approaches at each level and reveal regulatory blind spots that leave critical assets exposed. By offering a structured lens to align future AI technologies and governance with trustworthy data practices, we underscore the urgency of rethinking data protection for modern AI techniques and provide timely guidance for developers, researchers, and regulators alike.

cs.LG↗

arXiv2025

Robust Knowledge Editing via Explicit Reasoning Chains for Distractor-Resilient Multi-Hop QA

Large language models (LLMs) encode vast amounts of world knowledge but remain static once trained, making the timely integration of emerging facts prohibitively expensive via full retraining. Knowledge-editing techniques have thus emerged to inject or overwrite specific facts into LLMs, yet they either over-rely on superficial cues or incur complex, iterative pipelines that collapse under noisy, multi-hop conditions. We introduce Reason-KE, an end-to-end reasoning-chain-based editing framework that steers a pretrained LLM through four structured stages-fact acknowledgment, relevance determination, selective application, and final reasoning-to filter distractors in a single pass. Trained on MQuAKE-CF with up to four irrelevant facts, Reason-KE elevates Qwen2.5-7B's multi-hop QA accuracy to 90.2% while suffering merely a 6.3% drop under heavy distraction and <1% when answers are leaked. Our quantitative analysis confirms Reason-KE's resilience and efficiency, establishing a new state-of-the-art for reliable LLM knowledge updates.

cs.CL↗

arXiv2025

On the Evolution of Federated Post-Training Large Language Models: A Model Accessibility View

Federated Learning (FL) enables training models across decentralized data silos while preserving client data privacy. Recent research has explored efficient methods for post-training large language models (LLMs) within FL to address computational and communication challenges. While existing approaches often rely on access to LLMs' internal information, which is frequently restricted in real-world scenarios, an inference-only paradigm (black-box FedLLM) has emerged to address these limitations. This paper presents a comprehensive survey on federated tuning for LLMs. We propose a taxonomy categorizing existing studies along two axes: model access-based and parameter efficiency-based optimization. We classify FedLLM approaches into white-box, gray-box, and black-box techniques, highlighting representative methods within each category. We review emerging research treating LLMs as black-box inference APIs and discuss promising directions and open challenges for future research.

cs.LG↗

arXiv2025

CopyrightShield: Enhancing Diffusion Model Security against Copyright Infringement Attacks

Diffusion models have attracted significant attention due to its exceptional data generation capabilities in fields such as image synthesis. However, recent studies have shown that diffusion models are vulnerable to copyright infringement attacks, where attackers inject strategically modified non-infringing images into the training set, inducing the model to generate infringing content under the prompt of specific poisoned captions. To address this issue, we first propose a defense framework, CopyrightShield, to defend against the above attack. Specifically, we analyze the memorization mechanism of diffusion models and find that attacks exploit the model's overfitting to specific spatial positions and prompts, causing it to reproduce poisoned samples under backdoor triggers. Based on this, we propose a poisoned sample detection method using spatial masking and data attribution to quantify poisoning risk and accurately identify hidden backdoor samples. To further mitigate memorization of poisoned features, we introduce an adaptive optimization strategy that integrates a dynamic penalty term into the training loss, reducing reliance on infringing features while preserving generative performance. Experimental results demonstrate that CopyrightShield significantly improves poisoned sample detection performance across two attack scenarios, achieving average F1-scores of 0.665, retarding the First-Attack Epoch (FAE) of 115.2% and decreasing the Copyright Infringement Rate (CIR) by 56.7%. Compared to the SoTA backdoor defense in diffusion models, the defense effect is improved by about 25%, showcasing its superiority and practicality in enhancing the security of diffusion models.

cs.AI↗

arXiv2025

R1-VL: Learning to Reason with Multimodal Large Language Models via Step-wise Group Relative Policy Optimization

Recent studies generally enhance MLLMs' reasoning capabilities via supervised fine-tuning on high-quality chain-of-thought reasoning data, which often leads models to merely imitate successful reasoning paths without understanding what the wrong reasoning paths are. In this work, we aim to enhance the MLLMs' reasoning ability beyond passively imitating positive reasoning paths. To this end, we design Step-wise Group Relative Policy Optimization (StepGRPO), a new online reinforcement learning framework that enables MLLMs to self-improve reasoning ability via simple, effective and dense step-wise rewarding. Specifically, StepGRPO introduces two novel rule-based reasoning rewards: Step-wise Reasoning Accuracy Reward (StepRAR) and Step-wise Reasoning Validity Reward (StepRVR). StepRAR rewards the reasoning paths that contain necessary intermediate reasoning steps via a soft key-step matching technique, while StepRAR rewards reasoning paths that follow a well-structured and logically consistent reasoning process through a reasoning completeness and logic evaluation strategy. With the proposed StepGRPO, we introduce R1-VL, a series of MLLMs with outstanding capabilities in step-by-step reasoning. Extensive experiments over 8 benchmarks demonstrate the superiority of our methods.

cs.AI↗

arXiv2025

GLC++: Source-Free Universal Domain Adaptation through Global-Local Clustering and Contrastive Affinity Learning

Deep neural networks often exhibit sub-optimal performance under covariate and category shifts. Source-Free Domain Adaptation (SFDA) presents a promising solution to this dilemma, yet most SFDA approaches are restricted to closed-set scenarios. In this paper, we explore Source-Free Universal Domain Adaptation (SF-UniDA) aiming to accurately classify "known" data belonging to common categories and segregate them from target-private "unknown" data. We propose a novel Global and Local Clustering (GLC) technique, which comprises an adaptive one-vs-all global clustering algorithm to discern between target classes, complemented by a local k-NN clustering strategy to mitigate negative transfer. Despite the effectiveness, the inherent closed-set source architecture leads to uniform treatment of "unknown" data, impeding the identification of distinct "unknown" categories. To address this, we evolve GLC to GLC++, integrating a contrastive affinity learning strategy. We examine the superiority of GLC and GLC++ across multiple benchmarks and category shift scenarios. Remarkably, in the most challenging open-partial-set scenarios, GLC and GLC++ surpass GATE by 16.8\% and 18.9\% in H-score on VisDA, respectively. GLC++ enhances the novel category clustering accuracy of GLC by 4.1\% in open-set scenarios on Office-Home. Furthermore, the introduced contrastive learning strategy not only enhances GLC but also significantly facilitates existing methodologies. The code is available at https://github.com/ispc-lab/GLC-plus.

cs.CV↗

arXiv2025

Revisiting LLM Reasoning via Information Bottleneck

Large language models (LLMs) have recently demonstrated remarkable progress in reasoning capabilities through reinforcement learning with verifiable rewards (RLVR). By leveraging simple rule-based rewards, RL effectively incentivizes LLMs to produce extended chain-of-thought (CoT) reasoning trajectories, progressively guiding them toward correct answers. However, existing approaches remain largely heuristic and intuition-driven, limiting the development of principled methodologies. In this paper, we present a theoretical characterization of LLM reasoning grounded in information bottleneck (IB) principle, introducing IB-aware reasoning optimization (IBRO), a framework that encourages reasoning trajectories to be both informative about the final correct answer and generalizable across diverse prompts. We derive a practical token-level surrogate objective and propose an efficient approximation, resulting in the lightweight IB regularization method. This technique integrates seamlessly into existing RL-based post-training frameworks without additional computational overhead, requiring only a one-line code modification. Empirically, we validate IB regularization across multiple mathematical reasoning benchmarks and RL algorithms, demonstrating consistent improvements in LLM reasoning performance.

cs.AI↗

arXiv2025

Demonstration of Efficient Predictive Surrogates for Large-scale Quantum Processors

The ongoing development of quantum processors is driving breakthroughs in scientific discovery. Despite this progress, the formidable cost of fabricating large-scale quantum processors means they will remain rare for the foreseeable future, limiting their widespread application. To address this bottleneck, we introduce the concept of predictive surrogates, which are classical learning models designed to emulate the mean-value behavior of a given quantum processor with provably computational efficiency. In particular, we propose two predictive surrogates that can substantially reduce the need for quantum processor access in diverse practical scenarios. To demonstrate their potential in advancing digital quantum simulation, we use these surrogates to emulate a quantum processor with up to 20 programmable superconducting qubits, enabling efficient pre-training of variational quantum eigensolvers for families of transverse-field Ising models and identification of non-equilibrium Floquet symmetry-protected topological phases. Experimental results reveal that the predictive surrogates not only reduce measurement overhead by orders of magnitude, but can also surpass the performance of conventional, quantum-resource-intensive approaches. Collectively, these findings establish predictive surrogates as a practical pathway to broadening the impact of advanced quantum processors.

quant-ph↗

arXiv2025

ShadowCode: Towards (Automatic) External Prompt Injection Attack against Code LLMs

Recent advancements have led to the widespread adoption of code-oriented large language models (Code LLMs) for programming tasks. Despite their success in deployment, their security research is left far behind. This paper introduces a new attack paradigm: (automatic) external prompt injection against Code LLMs, where attackers generate concise, non-functional induced perturbations and inject them within a victim's code context. These induced perturbations can be disseminated through commonly used dependencies (e.g., packages or RAG's knowledge base), manipulating Code LLMs to achieve malicious objectives during the code completion process. Compared to existing attacks, this method is more realistic and threatening: it does not necessitate control over the model's training process, unlike backdoor attacks, and can achieve specific malicious objectives that are challenging for adversarial attacks. Furthermore, we propose ShadowCode, a simple yet effective method that automatically generates induced perturbations based on code simulation to achieve effective and stealthy external prompt injection. ShadowCode designs its perturbation optimization objectives by simulating realistic code contexts and employs a greedy optimization approach with two enhancement modules: forward reasoning enhancement and keyword-based perturbation design. We evaluate our method across 13 distinct malicious objectives, generating 31 threat cases spanning three popular programming languages. Our results demonstrate that ShadowCode successfully attacks three representative open-source Code LLMs (achieving up to a 97.9% attack success rate) and two mainstream commercial Code LLM-integrated applications (with over 90% attack success rate) across all threat cases, using only a 12-token non-functional induced perturbation. The code is available at https://github.com/LianPing-cyber/ShadowCodeEPI.

cs.CR↗