arXiv subjects

Huan Zhang

Publications and source records attributed to Huan Zhang.

At least 127 records · Page 7Linked to original sources

arXiv2025

Bridging Discrete and Continuous: A Multimodal Strategy for Complex Emotion Detection

In the domain of human-computer interaction, accurately recognizing and interpreting human emotions is crucial yet challenging due to the complexity and subtlety of emotional expressions. This study explores the potential for detecting a rich and flexible range of emotions through a multimodal approach which integrates facial expressions, voice tones, and transcript from video clips. We propose a novel framework that maps variety of emotions in a three-dimensional Valence-Arousal-Dominance (VAD) space, which could reflect the fluctuations and positivity/negativity of emotions to enable a more variety and comprehensive representation of emotional states. We employed K-means clustering to transit emotions from traditional discrete categorization to a continuous labeling system and built a classifier for emotion recognition upon this system. The effectiveness of the proposed model is evaluated using the MER2024 dataset, which contains culturally consistent video clips from Chinese movies and TV series, annotated with both discrete and open-vocabulary emotion labels. Our experiment successfully achieved the transformation between discrete and continuous models, and the proposed model generated a more diverse and comprehensive set of emotion vocabulary while maintaining strong accuracy.

cs.MM↗

arXiv2025

Perceive-IR: Learning to Perceive Degradation Better for All-in-One Image Restoration

Existing All-in-One image restoration methods often fail to perceive degradation types and severity levels simultaneously, overlooking the importance of fine-grained quality perception. Moreover, these methods often utilize highly customized backbones, which hinder their adaptability and integration into more advanced restoration networks. To address these limitations, we propose Perceive-IR, a novel backbone-agnostic All-in-One image restoration framework designed for fine-grained quality control across various degradation types and severity levels. Its modular structure allows core components to function independently of specific backbones, enabling seamless integration into advanced restoration models without significant modifications. Specifically, Perceive-IR operates in two key stages: 1) multi-level quality-driven prompt learning stage, where a fine-grained quality perceiver is meticulously trained to discern three tier quality levels by optimizing the alignment between prompts and images within the CLIP perception space. This stage ensures a nuanced understanding of image quality, laying the groundwork for subsequent restoration; 2) restoration stage, where the quality perceiver is seamlessly integrated with a difficulty-adaptive perceptual loss, forming a quality-aware learning strategy. This strategy not only dynamically differentiates sample learning difficulty but also achieves fine-grained quality control by driving the restored image toward the ground truth while pulling it away from both low- and medium-quality samples.

cs.CV↗

arXiv2025

Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks

Vision Language Models (VLMs) can produce unintended and harmful content when exposed to adversarial attacks, particularly because their vision capabilities create new vulnerabilities. Existing defenses, such as input preprocessing, adversarial training, and response evaluation-based methods, are often impractical for real-world deployment due to their high costs. To address this challenge, we propose ASTRA, an efficient and effective defense by adaptively steering models away from adversarial feature directions to resist VLM attacks. Our key procedures involve finding transferable steering vectors representing the direction of harmful response and applying adaptive activation steering to remove these directions at inference time. To create effective steering vectors, we randomly ablate the visual tokens from the adversarial images and identify those most strongly associated with jailbreaks. These tokens are then used to construct steering vectors. During inference, we perform the adaptive steering method that involves the projection between the steering vectors and calibrated activation, resulting in little performance drops on benign inputs while strongly avoiding harmful outputs under adversarial inputs. Extensive experiments across multiple models and baselines demonstrate our state-of-the-art performance and high efficiency in mitigating jailbreak risks. Additionally, ASTRA exhibits good transferability, defending against unseen attacks (i.e., structured-based attack, perturbation-based attack with project gradient descent variants, and text-only attack). Our code is available at \url{https://github.com/ASTRAL-Group/ASTRA}.

cs.CV↗

arXiv2025

From Aesthetics to Human Preferences: Comparative Perspectives of Evaluating Text-to-Music Systems

Evaluating generative models remains a fundamental challenge, particularly when the goal is to reflect human preferences. In this paper, we use music generation as a case study to investigate the gap between automatic evaluation metrics and human preferences. We conduct comparative experiments across five state-of-the-art music generation approaches, assessing both perceptual quality and distributional similarity to human-composed music. Specifically, we evaluate synthesis music from various perceptual dimensions and examine reference-based metrics such as Mauve Audio Divergence (MAD) and Kernel Audio Distance (KAD). Our findings reveal significant inconsistencies across the different metrics, highlighting the limitation of the current evaluation practice. To support further research, we release a benchmark dataset comprising samples from multiple models. This study provides a broader perspective on the alignment of human preference in generative modeling, advocating for more human-centered evaluation strategies across domains.

eess.AS↗

arXiv2025

Unveiling the Underwater World: CLIP Perception Model-Guided Underwater Image Enhancement

High-quality underwater images are essential for both machine vision tasks and viewers with their aesthetic appeal.However, the quality of underwater images is severely affected by light absorption and scattering. Deep learning-based methods for Underwater Image Enhancement (UIE) have achieved good performance. However, these methods often overlook considering human perception and lack sufficient constraints within the solution space. Consequently, the enhanced images often suffer from diminished perceptual quality or poor content restoration.To address these issues, we propose a UIE method with a Contrastive Language-Image Pre-Training (CLIP) perception loss module and curriculum contrastive regularization. Above all, to develop a perception model for underwater images that more aligns with human visual perception, the visual semantic feature extraction capability of the CLIP model is leveraged to learn an appropriate prompt pair to map and evaluate the quality of underwater images. This CLIP perception model is then incorporated as a perception loss module into the enhancement network to improve the perceptual quality of enhanced images. Furthermore, the CLIP perception model is integrated with the curriculum contrastive regularization to enhance the constraints imposed on the enhanced images within the CLIP perceptual space, mitigating the risk of both under-enhancement and over-enhancement. Specifically, the CLIP perception model is employed to assess and categorize the learning difficulty level of negatives in the regularization process, ensuring comprehensive and nuanced utilization of distorted images and negatives with varied quality levels. Extensive experiments demonstrate that our method outperforms state-of-the-art methods in terms of visual quality and generalization ability.

cs.CV↗

arXiv2025

Neural Contraction Metrics with Formal Guarantees for Discrete-Time Nonlinear Dynamical Systems

Contraction metrics are crucial in control theory because they provide a powerful framework for analyzing stability, robustness, and convergence of various dynamical systems. However, identifying these metrics for complex nonlinear systems remains an open challenge due to the lack of scalable and effective tools. This paper explores the approach of learning verifiable contraction metrics parametrized as neural networks (NNs) for discrete-time nonlinear dynamical systems. While prior works on formal verification of contraction metrics for general nonlinear systems have focused on convex optimization methods (e.g. linear matrix inequalities, etc) under the assumption of continuously differentiable dynamics, the growing prevalence of NN-based controllers, often utilizing ReLU activations, introduces challenges due to the non-smooth nature of the resulting closed-loop dynamics. To bridge this gap, we establish a new sufficient condition for establishing formal neural contraction metrics for general discrete-time nonlinear systems assuming only the continuity of the dynamics. We show that from a computational perspective, our sufficient condition can be efficiently verified using the state-of-the-art neural network verifier $α,\!β$-CROWN, which scales up non-convex neural network verification via novel integration of symbolic linear bound propagation and branch-and-bound. Built upon our analysis tool, we further develop a learning method for synthesizing neural contraction metrics from sampled data. Finally, our approach is validated through the successful synthesis and verification of NN contraction metrics for various nonlinear examples.

math.OC↗

arXiv2025

Stealthy Backdoor Attack in Self-Supervised Learning Vision Encoders for Large Vision Language Models

Self-supervised learning (SSL) vision encoders learn high-quality image representations and thus have become a vital part of developing vision modality of large vision language models (LVLMs). Due to the high cost of training such encoders, pre-trained encoders are widely shared and deployed into many LVLMs, which are security-critical or bear societal significance. Under this practical scenario, we reveal a new backdoor threat that significant visual hallucinations can be induced into these LVLMs by merely compromising vision encoders. Because of the sharing and reuse of these encoders, many downstream LVLMs may inherit backdoor behaviors from encoders, leading to widespread backdoors. In this work, we propose BadVision, the first method to exploit this vulnerability in SSL vision encoders for LVLMs with novel trigger optimization and backdoor learning techniques. We evaluate BadVision on two types of SSL encoders and LVLMs across eight benchmarks. We show that BadVision effectively drives the LVLMs to attacker-chosen hallucination with over 99% attack success rate, causing a 77.6% relative visual understanding error while maintaining the stealthiness. SoTA backdoor detection methods cannot detect our attack effectively.

cs.CV↗

arXiv2025

The Emperor's New Clothes in Benchmarking? A Rigorous Examination of Mitigation Strategies for LLM Benchmark Data Contamination

Benchmark Data Contamination (BDC)-the inclusion of benchmark testing samples in the training set-has raised increasing concerns in Large Language Model (LLM) evaluation, leading to falsely inflated performance estimates and undermining evaluation reliability. To address this, researchers have proposed various mitigation strategies to update existing benchmarks, including modifying original questions or generating new ones based on them. However, a rigorous examination of the effectiveness of these mitigation strategies remains lacking. In this paper, we design a systematic and controlled pipeline along with two novel metrics-fidelity and contamination resistance-to provide a fine-grained and comprehensive assessment of existing BDC mitigation strategies. Previous assessment methods, such as accuracy drop and accuracy matching, focus solely on aggregate accuracy, often leading to incomplete or misleading conclusions. Our metrics address this limitation by emphasizing question-level evaluation result matching. Extensive experiments with 10 LLMs, 5 benchmarks, 20 BDC mitigation strategies, and 2 contamination scenarios reveal that no existing strategy significantly improves resistance over the vanilla case (i.e., no benchmark update) across all benchmarks, and none effectively balances fidelity and contamination resistance. These findings underscore the urgent need for designing more effective BDC mitigation strategies. Our code repository is available at https://github.com/ASTRAL-Group/BDC_mitigation_assessment.

cs.AI↗

arXiv2025

BaB-ND: Long-Horizon Motion Planning with Branch-and-Bound and Neural Dynamics

Neural-network-based dynamics models learned from observational data have shown strong predictive capabilities for scene dynamics in robotic manipulation tasks. However, their inherent non-linearity presents significant challenges for effective planning. Current planning methods, often dependent on extensive sampling or local gradient descent, struggle with long-horizon motion planning tasks involving complex contact events. In this paper, we present a GPU-accelerated branch-and-bound (BaB) framework for motion planning in manipulation tasks that require trajectory optimization over neural dynamics models. Our approach employs a specialized branching heuristics to divide the search space into subdomains, and applies a modified bound propagation method, inspired by the state-of-the-art neural network verifier alpha-beta-CROWN, to efficiently estimate objective bounds within these subdomains. The branching process guides planning effectively, while the bounding process strategically reduces the search space. Our framework achieves superior planning performance, generating high-quality state-action trajectories and surpassing existing methods in challenging, contact-rich manipulation tasks such as non-prehensile planar pushing with obstacles, object sorting, and rope routing in both simulated and real-world settings. Furthermore, our framework supports various neural network architectures, ranging from simple multilayer perceptrons to advanced graph neural dynamics models, and scales efficiently with different model sizes.

cs.RO↗

arXiv2025

Eggly: Designing Mobile Augmented Reality Neurofeedback Training Games for Children with Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that affects how children communicate and relate to other people and the world around them. Emerging studies have shown that neurofeedback training (NFT) games are an effective and playful intervention to enhance social and attentional capabilities for autistic children. However, NFT is primarily available in a clinical setting that is hard to scale. Also, the intervention demands deliberately-designed gamified feedback with fun and enjoyment, where little knowledge has been acquired in the HCI community. Through a ten-month iterative design process with four domain experts, we developed Eggly, a mobile NFT game based on a consumer-grade EEG headband and a tablet. Eggly uses novel augmented reality (AR) techniques to offer engagement and personalization, enhancing their training experience. We conducted two field studies (a single-session study and a three-week multi-session study) with a total of five autistic children to assess Eggly in practice at a special education center. Both quantitative and qualitative results indicate the effectiveness of the approach as well as contribute to the design knowledge of creating mobile AR NFT games.

cs.HC↗

arXiv2025

Abstract Rendering: Computing All that is Seen in Gaussian Splat Scenes

We introduce abstract rendering, a method for computing a set of images by rendering a scene from a continuously varying range of camera positions. The resulting abstract image-which encodes an infinite collection of possible renderings-is represented using constraints on the image matrix, enabling rigorous uncertainty propagation through the rendering process. This capability is particularly valuable for the formal verification of vision-based autonomous systems and other safety-critical applications. Our approach operates on Gaussian splat scenes, an emerging representation in computer vision and robotics. We leverage efficient piecewise linear bound propagation to abstract fundamental rendering operations, while addressing key challenges that arise in matrix inversion and depth sorting-two operations not directly amenable to standard approximations. To handle these, we develop novel linear relational abstractions that maintain precision while ensuring computational efficiency. These abstractions not only power our abstract rendering algorithm but also provide broadly applicable tools for other rendering problems. Our implementation, AbstractSplat, is optimized for scalability, handling up to 750k Gaussians while allowing users to balance memory and runtime through tile and batch-based computation. Compared to the only existing abstract image method for mesh-based scenes, AbstractSplat achieves 2-14x speedups while preserving precision. Our results demonstrate that continuous camera motion, rotations, and scene variations can be rigorously analyzed at scale, making abstract rendering a powerful tool for uncertainty-aware vision applications.

cs.CV↗

arXiv2025

MIH-TCCT: Mitigating Inconsistent Hallucinations in LLMs via Event-Driven Text-Code Cyclic Training

Recent methodologies utilizing synthetic datasets have aimed to address inconsistent hallucinations in large language models (LLMs); however,these approaches are primarily tailored to specific tasks, limiting their generalizability. Inspired by the strong performance of code-trained models in logic-intensive domains, we propose a novel framework that leverages event-based text to generate corresponding code and employs cyclic training to transfer the logical consistency of code to natural language effectively. Our method significantly reduces inconsistent hallucinations across three leading LLMs and two categories of natural language tasks while maintaining overall performance. This framework effectively alleviates hallucinations without necessitating adaptation to downstream tasks, demonstrating generality and providing new perspectives to tackle the challenge of inconsistent hallucinations.

cs.AI↗

arXiv2025

DynaMath: A Dynamic Visual Benchmark for Evaluating Mathematical Reasoning Robustness of Vision Language Models

The rapid advancements in Vision-Language Models (VLMs) have shown great potential in tackling mathematical reasoning tasks that involve visual context. Unlike humans who can reliably apply solution steps to similar problems with minor modifications, we found that SOTA VLMs like GPT-4o can consistently fail in these scenarios, revealing limitations in their mathematical reasoning capabilities. In this paper, we investigate the mathematical reasoning robustness in VLMs and evaluate how well these models perform under different variants of the same question, such as changes in visual numerical values or function graphs. While several vision-based math benchmarks have been developed to assess VLMs' problem-solving capabilities, these benchmarks contain only static sets of problems and cannot easily evaluate mathematical reasoning robustness. To fill this gap, we introduce DynaMath, a dynamic visual math benchmark designed for in-depth assessment of VLMs. DynaMath includes 501 high-quality, multi-topic seed questions, each represented as a Python program. Those programs are carefully designed and annotated to enable the automatic generation of a much larger set of concrete questions, including many different types of visual and textual variations. DynaMath allows us to evaluate the generalization ability of VLMs, by assessing their performance under varying input conditions of a seed question. We evaluated 14 SOTA VLMs with 5,010 generated concrete questions. Our results show that the worst-case model accuracy, defined as the percentage of correctly answered seed questions in all 10 variants, is significantly lower than the average-case accuracy. Our analysis emphasizes the need to study the robustness of VLMs' reasoning abilities, and DynaMath provides valuable insights to guide the development of more reliable models for mathematical reasoning.

cs.CV↗

arXiv2025

RenderBox: Expressive Performance Rendering with Text Control

Expressive music performance rendering involves interpreting symbolic scores with variations in timing, dynamics, articulation, and instrument-specific techniques, resulting in performances that capture musical can emotional intent. We introduce RenderBox, a unified framework for text-and-score controlled audio performance generation across multiple instruments, applying coarse-level controls through natural language descriptions and granular-level controls using music scores. Based on a diffusion transformer architecture and cross-attention joint conditioning, we propose a curriculum-based paradigm that trains from plain synthesis to expressive performance, gradually incorporating controllable factors such as speed, mistakes, and style diversity. RenderBox achieves high performance compared to baseline models across key metrics such as FAD and CLAP, and also tempo and pitch accuracy under different prompting tasks. Subjective evaluation further demonstrates that RenderBox is able to generate controllable expressive performances that sound natural and musically engaging, aligning well with prompts and intent.

eess.AS↗

arXiv2025

Neural Network Verification with Branch-and-Bound for General Nonlinearities

Branch-and-bound (BaB) is among the most effective techniques for neural network (NN) verification. However, existing works on BaB for NN verification have mostly focused on NNs with piecewise linear activations, especially ReLU networks. In this paper, we develop a general framework, named GenBaB, to conduct BaB on general nonlinearities to verify NNs with general architectures, based on linear bound propagation for NN verification. To decide which neuron to branch, we design a new branching heuristic which leverages linear bounds as shortcuts to efficiently estimate the potential improvement after branching. To decide nontrivial branching points for general nonlinear functions, we propose to pre-optimize branching points, which can be efficiently leveraged during verification with a lookup table. We demonstrate the effectiveness of our GenBaB on verifying a wide range of NNs, including NNs with activation functions such as Sigmoid, Tanh, Sine and GeLU, as well as NNs involving multi-dimensional nonlinear operations such as multiplications in LSTMs and Vision Transformers. Our framework also allows the verification of general nonlinear computation graphs and enables verification applications beyond simple NNs, particularly for AC Optimal Power Flow (ACOPF). GenBaB is part of the latest $α$,$β$-CROWN, the winner of the 4th and the 5th International Verification of Neural Networks Competition (VNN-COMP 2023 and 2024). Code for reproducing the experiments is available at https://github.com/shizhouxing/GenBaB.

cs.LG↗

arXiv2025

Compensation based Dictionary Transfer for Similar Multispectral Image Spectral Super-resolution

Utilizing a spectral dictionary learned from a couple of similar-scene multi- and hyperspectral image, it is possible to reconstruct a desired hyperspectral image only with one single multispectral image. However, the differences between the similar scene and the desired hyperspectral image make it difficult to directly apply the spectral dictionary from the training domain to the task domain. To this end, a compensation matrix based dictionary transfer method for the similar-scene multispectral image spectral super-resolution is proposed in this paper, trying to reconstruct a more accurate high spatial resolution hyperspectral image. Specifically, a spectral dictionary transfer scheme is established by using a compensation matrix with similarity constraint, to transfer the spectral dictionary learned in the training domain to the spectral super-resolution domain. Subsequently, the sparse coefficient matrix is optimized under sparse and low-rank constraints. Experimental results on two AVIRIS datasets from different scenes indicate that, the proposed method outperforms other related SOTA methods.

eess.IV↗

arXiv2024

Verified Safe Reinforcement Learning for Neural Network Dynamic Models

Learning reliably safe autonomous control is one of the core problems in trustworthy autonomy. However, training a controller that can be formally verified to be safe remains a major challenge. We introduce a novel approach for learning verified safe control policies in nonlinear neural dynamical systems while maximizing overall performance. Our approach aims to achieve safety in the sense of finite-horizon reachability proofs, and is comprised of three key parts. The first is a novel curriculum learning scheme that iteratively increases the verified safe horizon. The second leverages the iterative nature of gradient-based learning to leverage incremental verification, reusing information from prior verification runs. Finally, we learn multiple verified initial-state-dependent controllers, an idea that is especially valuable for more complex domains where learning a single universal verified safe controller is extremely challenging. Our experiments on five safe control problems demonstrate that our trained controllers can achieve verified safety over horizons that are as much as an order of magnitude longer than state-of-the-art baselines, while maintaining high reward, as well as a perfect safety record over entire episodes. Our code is available at https://github.com/jlwu002/VSRL.

cs.LG↗

arXiv2024

Regularizing Hidden States Enables Learning Generalizable Reward Model for LLMs

Reward models trained on human preference data have been proven to effectively align Large Language Models (LLMs) with human intent within the framework of reinforcement learning from human feedback (RLHF). However, current reward models have limited generalization capabilities to unseen prompts and responses, which can lead to an unexpected phenomenon known as reward over-optimization, resulting in a decline in actual performance due to excessive optimization of rewards. While previous research has advocated for constraining policy optimization, our study introduces a novel approach to enhance the reward model's generalization ability against distribution shifts by regularizing the hidden states. Specifically, we retain the base model's language model head and incorporate a suite of text-generation losses to preserve the hidden states' text-generation capabilities, while concurrently learning a reward head behind the same hidden states. Our experimental results demonstrate that the introduced regularization technique markedly improves the accuracy of learned reward models across a variety of out-of-distribution (OOD) tasks and effectively alleviates the over-optimization issue in RLHF, offering a more reliable and robust preference learning paradigm.

cs.CL↗