arXiv subjects

Jiacheng Zhang

Publications and source records attributed to Jiacheng Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Sample-Specific Noise Injection For Diffusion-Based Adversarial Purification

Diffusion-based purification (DBP) methods aim to remove adversarial noise from the input sample by first injecting Gaussian noise through a forward diffusion process, and then recovering the clean example through a reverse generative process. In the above process, how much Gaussian noise is injected to the input sample is key to the success of DBP methods, which is controlled by a constant noise level $t^*$ for all samples in existing methods. In this paper, we discover that an optimal $t^*$ for each sample indeed could be different. Intuitively, the cleaner a sample is, the less the noise it should be injected, and vice versa. Motivated by this finding, we propose a new framework, called Sample-specific Score-aware Noise Injection (SSNI). Specifically, SSNI uses a pre-trained score network to estimate how much a data point deviates from the clean data distribution (i.e., score norms). Then, based on the magnitude of score norms, SSNI applies a reweighting function to adaptively adjust $t^*$ for each sample, achieving sample-specific noise injections. Empirically, incorporating our framework with existing DBP methods results in a notable improvement in both accuracy and robustness on CIFAR-10 and ImageNet-1K, highlighting the necessity to allocate distinct noise levels to different samples in DBP methods. Our code is available at: https://github.com/tmlr-group/SSNI.

cs.CV↗

arXiv2026

Semantic-aware Adversarial Fine-tuning for CLIP

Recent studies have shown that CLIP model's adversarial robustness in zero-shot classification tasks can be enhanced by adversarially fine-tuning its image encoder with adversarial examples (AEs), which are generated by minimizing the cosine similarity between images and a hand-crafted template (e.g., ''A photo of a {label}''). However, it has been shown that the cosine similarity between a single image and a single hand-crafted template is insufficient to measure the similarity for image-text pairs. Building on this, in this paper, we find that the AEs generated using cosine similarity may fail to fool CLIP when the similarity metric is replaced with semantically enriched alternatives, making the image encoder fine-tuned with these AEs less robust. To overcome this issue, we first propose a semantic-ensemble attack to generate semantic-aware AEs by minimizing the average similarity between the original image and an ensemble of refined textual descriptions. These descriptions are initially generated by a foundation model to capture core semantic features beyond hand-crafted templates and are then refined to reduce hallucinations. To this end, we propose Semantic-aware Adversarial Fine-Tuning (SAFT), which fine-tunes CLIP's image encoder with semantic-aware AEs. Extensive experiments show that SAFT outperforms current methods, achieving substantial improvements in zero-shot adversarial robustness across 16 datasets. Our code is available at: https://github.com/tmlr-group/SAFT.

cs.CV↗

arXiv2026

FeatureBench: Benchmarking Agentic Coding for Complex Feature Development

Agents powered by large language models (LLMs) are increasingly adopted in the software industry, contributing code as collaborators or even autonomous developers. As their presence grows, it becomes important to assess the current boundaries of their coding abilities. Existing agentic coding benchmarks, however, cover a limited task scope, e.g., bug fixing within a single pull request (PR), and often rely on non-executable evaluations or lack an automated approach for continually updating the evaluation coverage. To address such issues, we propose FeatureBench, a benchmark designed to evaluate agentic coding performance in end-to-end, feature-oriented software development. FeatureBench incorporates an execution-based evaluation protocol and a scalable test-driven method that automatically derives tasks from code repositories with minimal human effort. By tracing from unit tests along a dependency graph, our approach can identify feature-level coding tasks spanning multiple commits and PRs scattered across the development timeline, while ensuring the proper functioning of other features after the separation. Using this framework, we curated 200 challenging evaluation tasks and 3825 executable environments from 24 open-source repositories in the first version of our benchmark. Empirical evaluation reveals that the state-of-the-art agentic model, such as Claude 4.5 Opus, which achieves a 74.4% resolved rate on SWE-bench, succeeds on only 11.0% of tasks, opening new opportunities for advancing agentic coding. Moreover, benefiting from our automated task collection toolkit, FeatureBench can be easily scaled and updated over time to mitigate data leakage. The inherent verifiability of constructed environments also makes our method potentially valuable for agent training.

cs.SE↗

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

Let's Roll a BiFTA: Bi-refinement for Fine-grained Text-visual Alignment in Vision-Language Models

Recent research has shown that aligning fine-grained text descriptions with localized image patches can significantly improve the zero-shot performance of pre-trained vision-language models (e.g., CLIP). However, we find that both fine-grained text descriptions and localized image patches often contain redundant information, making text-visual alignment less effective. In this paper, we tackle this issue from two perspectives: \emph{View Refinement} and \emph{Description refinement}, termed as \textit{\textbf{Bi}-refinement for \textbf{F}ine-grained \textbf{T}ext-visual \textbf{A}lignment} (BiFTA). \emph{View refinement} removes redundant image patches with high \emph{Intersection over Union} (IoU) ratios, resulting in more distinctive visual samples. \emph{Description refinement} removes redundant text descriptions with high pairwise cosine similarity, ensuring greater diversity in the remaining descriptions. BiFTA achieves superior zero-shot performance on 6 benchmark datasets for both ViT-based and ResNet-based CLIP, justifying the necessity to remove redundant information in visual-text alignment.

cs.CV↗

arXiv2025

OnlineVPO: Align Video Diffusion Model with Online Video-Centric Preference Optimization

Video diffusion models (VDMs) have demonstrated remarkable capabilities in text-to-video (T2V) generation. Despite their success, VDMs still suffer from degraded image quality and flickering artifacts. To address these issues, some approaches have introduced preference learning to exploit human feedback to enhance the video generation. However, these methods primarily adopt the routine in the image domain without an in-depth investigation into video-specific preference optimization. In this paper, we reexamine the design of the video preference learning from two key aspects: feedback source and feedback tuning methodology, and present OnlineVPO, a more efficient preference learning framework tailored specifically for VDMs. On the feedback source, we found that the image-level reward model commonly used in existing methods fails to provide a human-aligned video preference signal due to the modality gap. In contrast, video quality assessment (VQA) models show superior alignment with human perception of video quality. Building on this insight, we propose leveraging VQA models as a proxy of humans to provide more modality-aligned feedback for VDMs. Regarding the preference tuning methodology, we introduce an online DPO algorithm tailored for VDMs. It not only enjoys the benefits of superior scalability in optimizing videos with higher resolution and longer duration compared with the existing method, but also mitigates the insufficient optimization issue caused by off-policy learning via online preference generation and curriculum preference update designs. Extensive experiments on the open-source video-diffusion model demonstrate OnlineVPO as a simple yet effective and, more importantly, scalable preference learning algorithm for video diffusion models.

cs.CV↗

arXiv2025

Dimension-free error estimate for diffusion model and optimal scheduling

Diffusion generative models have emerged as powerful tools for producing synthetic data from an empirically observed distribution. A common approach involves simulating the time-reversal of an Ornstein-Uhlenbeck (OU) process initialized at the true data distribution. Since the score function associated with the OU process is typically unknown, it is approximated using a trained neural network. This approximation, along with finite time simulation, time discretization and statistical approximation, introduce several sources of error whose impact on the generated samples must be carefully understood. Previous analyses have quantified the error between the generated and the true data distributions in terms of Wasserstein distance or Kullback-Leibler (KL) divergence. However, both metrics present limitations: KL divergence requires absolute continuity between distributions, while Wasserstein distance, though more general, leads to error bounds that scale poorly with dimension, rendering them impractical in high-dimensional settings. In this work, we derive an explicit, dimension-free bound on the discrepancy between the generated and the true data distributions. The bound is expressed in terms of a smooth test functional with bounded first and second derivatives. The key novelty lies in the use of this weaker, functional metric to obtain dimension-independent guarantees, at the cost of higher regularity on the test functions. As an application, we formulate and solve a variational problem to minimize the time-discretization error, leading to the derivation of an optimal time-scheduling strategy for the reverse-time diffusion. Interestingly, this scheduler has appeared previously in the literature in a different context; our analysis provides a new justification for its optimality, now grounded in minimizing the discretization bias in generative sampling.

stat.ML↗

arXiv2025

EventHallusion: Diagnosing Event Hallucinations in Video LLMs

Recently, Multimodal Large Language Models (MLLMs) have made significant progress in the video comprehension field. Despite remarkable content reasoning and instruction following capabilities they demonstrated, the hallucination problem of these VideoLLMs is less explored compared with its counterpart in the image domain. To mitigate this gap, we propose EventHallusion, a novel benchmark that focuses on assessing the VideoLLMs' hallucination toward event, the crux of video analysis. From a hallucination attribution perspective, our EventHallusion benchmark is curated to assess a VideoLLM's susceptibility toward language priors and vision-language biases. On the other hand, we also propose a simple yet effective method, called Temporal Contrastive Decoding (TCD), to tackle the hallucination problems of VideoLLMs. The proposed TCD method rectifies the model's bias toward its priors during the decoding stage by comparing the original video with a modified version, in which temporal cues are disrupted. Through comprehensive evaluation of eight open-source and two closed-source VideoLLMs on the proposed EventHallusion benchmark, we observe that the open-source models suffer significantly from hallucination problems, whereas the closed-source ones perform markedly better. By further equipping open-source VideoLLMs with the proposed TCD approach, evident performance improvements are achieved across most metrics in the EventHallusion benchmark. Our codes and benchmark data are available at https://github.com/Stevetich/EventHallusion.

cs.CV↗

arXiv2025

Deep Signature Approach for McKean-Vlasov FBSDEs in a Random Environment

Mean-field games with common noise provide a powerful framework for modeling the collective behavior of large populations subject to shared randomness, such as systemic risk in finance or environmental shocks in economics. These problems can be reformulated as McKean-Vlasov forward-backward stochastic differential equations (MV-FBSDEs) in a random environment, where the coefficients depend on the conditional law of the state given the common noise. Existing numerical methods, however, are largely limited to cases where interactions depend only on expectations or low-order moments, and therefore cannot address the general setting of full distributional dependence. In this work, we introduce a deep learning-based algorithm for solving MV-FBSDEs with common noise and general mean-field interactions. Building on fictitious play, our method iteratively solves conditional FBSDEs with fixed distributions, where the conditional law is efficiently represented using signatures, and then updates the distribution through supervised learning. Deep neural networks are employed both to solve the conditional FBSDEs and to approximate the distribution-dependent coefficients, enabling scalability to high-dimensional problems. Under suitable assumptions, we establish convergence in terms of the fictitious play iterations, with error controlled by the supervised learning step. Numerical experiments, including a distribution-dependent mean-field game with common noise, demonstrate the effectiveness of the proposed approach.

math.OC↗

arXiv2025

On time-consistent equilibrium stopping under aggregation of diverse discount rates

This paper studies a central planner's decision making on behalf of a group of members with diverse discount rates. In the context of optimal stopping, we work with an aggregation preference to incorporate all discount rates via an attitude function that reflects the aggregation rule chosen by the central planner. The problem formulation is also applicable to single agent's stopping problem with uncertain discount rate, where our aggregation preference coincides with the conventional smooth ambiguity preference. The resulting optimal stopping problem is time inconsistent, for which we develop an iterative approach using consistent planning and characterize all time-consistent mild equilibria as fixed points of an operator in the setting of one-dimensional diffusion processes. We provide some sufficient conditions on the underlying models and the attitude function such that the smallest mild equilibrium attains the optimal equilibrium. In addition, we show that the optimal equilibrium is a weak equilibrium. When the sufficient condition of the attitude function is violated, we illustrate by various examples that the characterization of the optimal equilibrium may differ significantly from some existing results for a single agent, which now sensitively depends on the attitude function and the diversity distribution of discount rates within the group.

q-fin.MF↗

arXiv2025

MoRA: On-the-fly Molecule-aware Low-Rank Adaptation Framework for LLM-based Multi-Modal Molecular Assistant

Effectively integrating molecular graph structures with Large Language Models (LLMs) is a key challenge in drug discovery. Most existing multi-modal alignment methods typically process these structures by fine-tuning the LLM or adding a static adapter simultaneously. However, these approaches have two main limitations: (1) it optimizes a shared parameter space across all molecular inputs, limiting the model's ability to capture instance-specific structural features; and (2) fine-tuning the LLM for molecular tasks can lead to catastrophic forgetting, undermining its general reasoning capabilities. In this paper, instead of static task-oriented adaptation, we propose an instance-specific parameter space alignment approach for each molecule on-the-fly. To this end, we introduce Molecule-aware Low-Rank Adaptation (MoRA) that produces a unique set of low-rank adaptation weights for each input molecular graph. These weights are then dynamically injected into a frozen LLM, allowing the model to adapt its reasoning to the structure of each molecular input, while preserving the LLM's core knowledge. Extensive experiments demonstrate that on key molecular tasks, such as chemical reaction prediction and molecular captioning, MoRA's instance-specific dynamic adaptation outperforms statically adapted baselines, including a 14.1% relative improvement in reaction prediction exact match and a 22% reduction in error for quantum property prediction. The code is available at https://github.com/jk-sounds/MoRA.

cs.LG↗

arXiv2025

Major-Minor Mean Field Game of Stopping: An Entropy Regularization Approach

This paper studies a discrete-time major-minor mean field game of stopping where the major player can choose either an optimal control or stopping time. We look for the relaxed equilibrium as a randomized stopping policy, which is formulated as a fixed point of a set-valued mapping, whose existence is challenging by direct arguments. To overcome the difficulties caused by the presence of a major player, we propose to study an auxiliary problem by considering entropy regularization in the major player's problem while formulating the minor players' optimal stopping problems as linear programming over occupation measures. We first show the existence of regularized equilibria as fixed points of some simplified set-valued operator using the Kakutani-Fan-Glicksberg fixed-point theorem. Next, we prove that the regularized equilibrium converges as the regularization parameter $λ$ tends to 0, and the limit corresponds to a fixed point of the original operator, thereby confirming the existence of a relaxed equilibrium in the original mean field game problem. We also extend this entropy regularization method to the mean-field game problem where the minor players choose optimal controls.

math.OC↗

arXiv2025

ReCode: Improving LLM-based Code Repair with Fine-Grained Retrieval-Augmented Generation

Recent advances in large language models (LLMs) have demonstrated impressive capabilities in code-related tasks, such as code generation and automated program repair. Despite their promising performance, most existing approaches for code repair suffer from high training costs or computationally expensive inference. Retrieval-augmented generation (RAG), with its efficient in-context learning paradigm, offers a more scalable alternative. However, conventional retrieval strategies, which are often based on holistic code-text embeddings, fail to capture the structural intricacies of code, resulting in suboptimal retrieval quality. To address the above limitations, we propose ReCode, a fine-grained retrieval-augmented in-context learning framework designed for accurate and efficient code repair. Specifically, ReCode introduces two key innovations: (1) an algorithm-aware retrieval strategy that narrows the search space using preliminary algorithm type predictions; and (2) a modular dual-encoder architecture that separately processes code and textual inputs, enabling fine-grained semantic matching between input and retrieved contexts. Furthermore, we propose RACodeBench, a new benchmark constructed from real-world user-submitted buggy code, which addresses the limitations of synthetic benchmarks and supports realistic evaluation. Experimental results on RACodeBench and competitive programming datasets demonstrate that ReCode achieves higher repair accuracy with significantly reduced inference cost, highlighting its practical value for real-world code repair scenarios.

cs.SE↗

arXiv2025

Ito's formula for flows of conditional measures on semimartingales

Motivated by recent development of mean-field systems with common noise, this paper establishes Ito's formula for flows of conditional probability measures under a common filtration associated with general semimartingales. This generalizes existing works on flows of conditional measures on Ito processes and flows of deterministic measure on general semimartingales. The key technical components involve constructing conditional independent copies and establishing the equivalence between stochastic integrals with respect to the conditional law of semimartingales and the conditional expectation of stochastic integrals with respect to copies of semimartingales. Ito's formula is then established for cylindrical functions through conditional independent copies, and extended to the general case through function approximations.

math.PR↗

arXiv2025

One Stone, Two Birds: Enhancing Adversarial Defense Through the Lens of Distributional Discrepancy

Statistical adversarial data detection (SADD) detects whether an upcoming batch contains adversarial examples (AEs) by measuring the distributional discrepancies between clean examples (CEs) and AEs. In this paper, we explore the strength of SADD-based methods by theoretically showing that minimizing distributional discrepancy can help reduce the expected loss on AEs. Despite these advantages, SADD-based methods have a potential limitation: they discard inputs that are detected as AEs, leading to the loss of useful information within those inputs. To address this limitation, we propose a two-pronged adversarial defense method, named Distributional-discrepancy-based Adversarial Defense (DAD). In the training phase, DAD first optimizes the test power of the maximum mean discrepancy (MMD) to derive MMD-OPT, which is a stone that kills two birds. MMD-OPT first serves as a guiding signal to minimize the distributional discrepancy between CEs and AEs to train a denoiser. Then, it serves as a discriminator to differentiate CEs and AEs during inference. Overall, in the inference stage, DAD consists of a two-pronged process: (1) directly feeding the detected CEs into the classifier, and (2) removing noise from the detected AEs by the distributional-discrepancy-based denoiser. Extensive experiments show that DAD outperforms current state-of-the-art (SOTA) defense methods by simultaneously improving clean and robust accuracy on CIFAR-10 and ImageNet-1K against adaptive white-box attacks. Codes are publicly available at: https://github.com/tmlr-group/DAD.

cs.LG↗

arXiv2025

Continuous-time mean field games: a primal-dual characterization

This paper establishes a primal-dual formulation for continuous-time mean field games (MFGs) and provides a complete analytical characterization of the set of all Nash equilibria (NEs). We first show that for any given mean field flow, the representative player's control problem with {\it measurable coefficients} is equivalent to a linear program over the space of occupation measures. We then establish the dual formulation of this linear program as a maximization problem over smooth subsolutions of the associated Hamilton-Jacobi-Bellman (HJB) equation, which plays a fundamental role in characterizing NEs of MFGs. Finally, a complete characterization of \emph{all NEs for MFGs} is established by the strong duality between the linear program and its dual problem. This strong duality is obtained by studying the solvability of the dual problem, and in particular through analyzing the regularity of the associated HJB equation. Compared with existing approaches for MFGs, the primal-dual formulation and its NE characterization do not require the convexity of the associated Hamiltonian or the uniqueness of its optimizer, and remain applicable when the HJB equation lacks classical or even continuous solutions.

math.OC↗

arXiv2025

BIFROST: Reinventing WiFi Signals Based on Dispersion Effect for Accurate Indoor Localization

WiFi-based device localization is a key enabling technology for smart applications, which has attracted numerous research studies in the past decade. Most of the existing approaches rely on Line-of-Sight (LoS) signals to work, while a critical problem is often neglected: In the real-world indoor environments, WiFi signals are everywhere, but very few of them are usable for accurate localization. As a result, the localization accuracy in practice is far from being satisfactory. This paper presents BIFROST, a novel hardware-software co-design for accurate indoor localization. The core idea of BIFROST is to reinvent WiFi signals, so as to provide sufficient LoS signals for localization. This is realized by exploiting the dispersion effect of signals emitted by the leaky wave antenna (LWA). We present a low-cost plug-in design of LWA that can generate orthogonal polarized signals: On one hand, LWA disperses signals of different frequencies to different angles, thus providing Angle-of-Arrival (AoA) information for the localized target. On the other hand, the target further leverages the antenna polarization mismatch to distinguish AoAs from different LWAs. In the software layer, fine-grained information in Channel State Information (CSI) is exploited to cope with multipath and noise. We implement BIFROST and evaluate its performance under various settings. The results show that the median localization error of BIFROST is 0.81m, which is 52.35% less than that of SpotFi, a state-of-the-art approach. SpotFi, when combined with BIFROST to work in the realistic settings, can reduce the localization error by 33.54%.

cs.ET↗

arXiv2025

QuinID: Enabling FDMA-Based Fully Parallel RFID with Frequency-Selective Antenna

Parallelizing passive Radio Frequency Identification (RFID) reading is an arguably crucial, yet unsolved challenge in modern IoT applications. Existing approaches remain limited to time-division operations and fail to read multiple tags simultaneously. In this paper, we introduce QuinID, the first frequency-division multiple access (FDMA) RFID system to achieve fully parallel reading. We innovatively exploit the frequency selectivity of the tag antenna rather than a conventional digital FDMA, bypassing the power and circuitry constraint of RFID tags. Specifically, we delicately design the frequency-selective antenna based on surface acoustic wave (SAW) components to achieve extreme narrow-band response, so that QuinID tags (i.e., QuinTags) operate exclusively within their designated frequency bands. By carefully designing the matching network and canceling various interference, a customized QuinReader communicates simultaneously with multiple QuinTags across distinct bands. QuinID maintains high compatibility with commercial RFID systems and presents a tag cost of less than 10 cents. We implement a 5-band QuinID system and evaluate its performance under various settings. The results demonstrate a fivefold increase in read rate, reaching up to 5000 reads per second.

cs.NI↗