arXiv subjects

Dongqi Han

Publications and source records attributed to Dongqi Han.

At least 19 recordsLinked to original sources

arXiv2026

Robust Harmful Features Under Jailbreak Attacks: Mechanistic Evidence from Attention Head Specialization in Large Language Models

Jailbreak attacks bypass LLM safety alignment, yet their mechanisms remain poorly understood. We provide evidence that attacks do not comprehensively eliminate safety features, but instead selectively suppress specific attention heads. We identify two functionally differentiated types: Adversarially Compromised Heads (ACHs) concentrated in early layers, which are suppressed under attacks, and Safety-Aligned Heads (SAHs) in mid-layers, which maintain robust activations even when attacks succeed. Ablation studies support the causal role of ACHs and the contribution of SAHs to robust activations: suppressing a small number of ACHs is sufficient to induce jailbreak-like behavior on normally refused inputs, while removing SAHs substantially weakens mid-layer safety activations. Token-level attribution further shows that ACH suppression is driven specifically by attack-template tokens, providing a mechanistic account of why attacks can bypass refusal decisions through ACH suppression while leaving internal safety signals sustained by SAHs -- a phenomenon we term Robust Harmful Features. To validate the practical significance of this robustness, we show that simply reading these persistent activations -- without any training -- yields competitive aggregate detection performance with strong adversarial robustness.

cs.CR↗

arXiv2026

Kuramoto Oscillatory Phase Encoding: Neuro-inspired Synchronization for Improved Learning Efficiency

Spatiotemporal neural dynamics and oscillatory synchronization are widely implicated in biological information processing and have been hypothesized to support flexible coordination such as feature binding. By contrast, most deep learning architectures represent and propagate information through activation values, neglecting the joint dynamics of rate and phase. In this work, we introduce Kuramoto oscillatory Phase Encoding (KoPE) as an additional, evolving phase state to Vision Transformers, incorporating a neuro-inspired synchronization mechanism to advance learning efficiency. We show that KoPE can improve training, parameter, and data efficiency of vision models through synchronization-enhanced structure learning. Moreover, KoPE benefits tasks requiring structured understanding, including semantic and panoptic segmentation, representation alignment with language, and few-shot abstract visual reasoning (ARC-AGI). Theoretical analysis and empirical verification further suggest that KoPE can accelerate attention concentration for learning efficiency. These results indicate that synchronization can serve as a scalable, neuro-inspired mechanism for advancing state-of-the-art neural network models. Code is avaliable at https://github.com/microsoft/Neuro-inspired_Phase_Encoding.

cs.LG↗

arXiv2026

Improving Diffusion Planners by Self-Supervised Action Gating with Energies

Diffusion planners are a strong approach for offline reinforcement learning, but they can fail when value-guided selection favours trajectories that score well yet are locally inconsistent with the environment dynamics, resulting in brittle execution. We propose Self-supervised Action Gating with Energies (SAGE), an inference-time re-ranking method that penalises dynamically inconsistent plans using a latent consistency signal. SAGE trains a Joint-Embedding Predictive Architecture (JEPA) encoder on offline state sequences and an action-conditioned latent predictor for short horizon transitions. At test time, SAGE assigns each sampled candidate an energy given by its latent prediction error and combines this feasibility score with value estimates to select actions. SAGE can integrate into existing diffusion planning pipelines that can sample trajectories and select actions via value scoring; it requires no environment rollouts and no policy re-training. Across locomotion, navigation, and manipulation benchmarks, SAGE improves the performance and robustness of diffusion planners.

cs.LG↗

arXiv2026

State-Action Inpainting Diffuser for Continuous Control with Delay

Signal delay poses a fundamental challenge in continuous control and reinforcement learning (RL) by introducing a temporal gap between interaction and perception. Current solutions have largely evolved along two distinct paradigms: model-free approaches which utilize state augmentation to preserve Markovian properties, and model-based methods which focus on inferring latent beliefs via dynamics modeling. In this paper, we bridge these perspectives by introducing State-Action Inpainting Diffuser (SAID), a framework that integrates the inductive bias of dynamics learning with the direct decision-making capability of policy optimization. By formulating the problem as a joint sequence inpainting task, SAID implicitly captures environmental dynamics while directly generating consistent plans, effectively operating at the intersection of model-based and model-free paradigms. Crucially, this generative formulation allows SAID to be seamlessly applied to both online and offline RL. Extensive experiments on delayed continuous control benchmarks demonstrate that SAID achieves state-of-the-art and robust performance. Our study suggests a new methodology to advance the field of RL with delay.

cs.AI↗

arXiv2025

EEGChaT: A Transformer-Based Modular Channel Selector for SEEG Analysis

Analyzing stereoelectroencephalography (SEEG) signals is critical for brain-computer interface (BCI) applications and neuroscience research, yet poses significant challenges due to the large number of input channels and their heterogeneous relevance. Traditional channel selection methods struggle to scale or provide meaningful interpretability for SEEG data. In this work, we propose EEGChaT, a novel Transformer-based channel selection module designed to automatically identify the most task-relevant channels in SEEG recordings. EEGChaT introduces Channel Aggregation Tokens (CATs) to aggregate information across channels, and leverages an improved Attention Rollout technique to compute interpretable, quantitative channel importance scores. We evaluate EEGChaT on the DuIN dataset, demonstrating that integrating EEGChaT with existing classification models consistently improves decoding accuracy, achieving up to 17\% absolute gains. Furthermore, the channel weights produced by EEGChaT show substantial overlap with manually selected channels, supporting the interpretability of the approach. Our results suggest that EEGChaT is an effective and generalizable solution for channel selection in high-dimensional SEEG analysis, offering both enhanced performance and insights into neural signal relevance.

cs.LG↗

arXiv2025

VidGuard-R1: AI-Generated Video Detection and Explanation via Reasoning MLLMs and RL

The rapid proliferation of AI-generated video necessitates robust detection tools that offer both high accuracy and human-interpretable explanations. While existing MLLM-based detectors rely on supervised fine-tuning (SFT) or direct preference optimization (DPO), these methods are often bottlenecked by static, pre-labeled datasets that fail to capture the evolving, multi-step physical inconsistencies of modern generative models. To bridge this gap, we introduce VidGuard-R1, the first video authenticity detector to utilize group relative policy optimization (GRPO). Moving beyond passive preference matching, VidGuard-R1 employs a reinforcement learning framework that encourages the model to explore and rank multiple reasoning paths. By introducing specialized reward models for temporal stability and diffusion-aware complexity, we incentivize the model to discover 'physics-grounded' artifacts. Our contributions include: (1) a curated dataset of 140,000 challenging real/fake video pairs; (2) a GRPO-based training paradigm that achieves state-of-the-art zero-shot performance; and (3) a reasoning-first architecture that provides precise, verifiable rationales for its forensic judgments. Project website: https://vidguard-r1.github.io/.

cs.CV↗

arXiv2025

Mind the Gap: A Decade-Scale Empirical Study of Multi-Stakeholder Dynamics in VR Ecosystem

In the development and evolution of VR ecosystem, platform stakeholders continuously adapt their products in response to user and technical feedback, often reflected in subtle shifts in discussion topics or system updates. A comprehensive understanding of these changes is essential for identifying gaps between user expectations and developer actions, which can guide more effective quality assurance and user-centered innovation. While previous studies have analyzed either user reviews or developer discussions in isolation, such approaches typically fail to reveal how specific user concerns are (or are not) addressed by corresponding technical activities. To address this limitation, our study introduces a multi-view empirical framework that systematically compares and aligns stakeholder perspectives. By applying topic modeling and quantitative impact analysis to 944,320 user reviews and 389,477 developer posts, we identify not only the overlap in concerns (e.g., performance, input methods), but also clear gaps in areas like inclusivity and community safety (e.g., LGBTQ+ representation, child-friendly content). Our findings show that while users repeatedly raise such issues, they are rarely discussed in developer forums. These insights enable data-driven recommendations for closing the user-developer gap in VR ecosystems, offering practical implications for platform governance and the design of next-generation VR systems.

cs.SE↗

arXiv2025

EgoBrain: Synergizing Minds and Eyes For Human Action Understanding

The integration of brain-computer interfaces (BCIs), in particular electroencephalography (EEG), with artificial intelligence (AI) has shown tremendous promise in decoding human cognition and behavior from neural signals. In particular, the rise of multimodal AI models have brought new possibilities that have never been imagined before. Here, we present EgoBrain -- the world's first large-scale, temporally aligned multimodal dataset that synchronizes first-person (egocentric) vision and EEG of human brain over extended periods of time, establishing a new paradigm for human-centered behavior analysis. This dataset comprises 61 hours of synchronized 32-channel EEG recordings and first-person video from 40 participants engaged in 29 categories of daily activities. We then developed a multimodal learning framework to fuse EEG and vision for action understanding, validated across both cross-subject and cross-environment challenges, achieving an action recognition accuracy of 66.70%. EgoBrain paves the way toward a unified framework for multimodal and egocentric brain-computer interfaces, bridging neural signals and first-person perception. Our dataset and code are publicly available at: https://huggingface.co/datasets/ut-vision/EgoBrain and https://github.com/ut-vision/EgoBrain .

cs.AI↗

arXiv2025

Language of Network: A Generative Pre-trained Model for Encrypted Traffic Comprehension

The increasing demand for privacy protection and security considerations leads to a significant rise in the proportion of encrypted network traffic. Since traffic content becomes unrecognizable after encryption, accurate analysis is challenging, making it difficult to classify applications and detect attacks. Deep learning is currently the predominant approach for encrypted traffic classification through feature analysis. However, these methods face limitations due to their high dependence on labeled data and difficulties in detecting attack variants. First, their performance is highly sensitive to data quality, where the highcost manual labeling process and dataset imbalance significantly degrade results. Second, the rapid evolution of attack patterns makes it challenging for models to identify new types of attacks. To tackle these challenges, we present GBC, a generative model based on pre-training for encrypted traffic comprehension. Since traditional tokenization methods are primarily designed for natural language, we propose a protocol-aware tokenization approach for encrypted traffic that improves model comprehension of fields specific to network traffic. In addition, GBC employs pretraining to learn general representations from extensive unlabeled traffic data. Through prompt learning, it effectively adapts to various downstream tasks, enabling both high-quality traffic generation and effective detection. Evaluations across multiple datasets demonstrate that GBC achieves superior results in both traffic classification and generation tasks, resulting in a 5% improvement in F1 score compared to state-of-the-art methods for classification tasks.

cs.CR↗

arXiv2025

Do Not Let Low-Probability Tokens Over-Dominate in RL for LLMs

Reinforcement learning (RL) has become a cornerstone for enhancing the reasoning capabilities of large language models (LLMs), with recent innovations such as Group Relative Policy Optimization (GRPO) demonstrating exceptional effectiveness. In this study, we identify a critical yet underexplored issue in RL training: low-probability tokens disproportionately influence model updates due to their large gradient magnitudes. This dominance hinders the effective learning of high-probability tokens, whose gradients are essential for LLMs' performance but are substantially suppressed. To mitigate this interference, we propose two novel methods: Advantage Reweighting and Low-Probability Token Isolation (Lopti), both of which effectively attenuate gradients from low-probability tokens while emphasizing parameter updates driven by high-probability tokens. Our approaches promote balanced updates across tokens with varying probabilities, thereby enhancing the efficiency of RL training. Experimental results demonstrate that they substantially improve the performance of GRPO-trained LLMs, achieving up to a 46.2% improvement in K&K Logic Puzzle reasoning tasks. Our implementation is available at https://github.com/zhyang2226/AR-Lopti.

cs.CL↗

arXiv2025

What Makes a Good Diffusion Planner for Decision Making?

Diffusion models have recently shown significant potential in solving decision-making problems, particularly in generating behavior plans -- also known as diffusion planning. While numerous studies have demonstrated the impressive performance of diffusion planning, the mechanisms behind the key components of a good diffusion planner remain unclear and the design choices are highly inconsistent in existing studies. In this work, we address this issue through systematic empirical experiments on diffusion planning in an offline reinforcement learning (RL) setting, providing practical insights into the essential components of diffusion planning. We trained and evaluated over 6,000 diffusion models, identifying the critical components such as guided sampling, network architecture, action generation and planning strategy. We revealed that some design choices opposite to the common practice in previous work in diffusion planning actually lead to better performance, e.g., unconditional sampling with selection can be better than guided sampling and Transformer outperforms U-Net as denoising network. Based on these insights, we suggest a simple yet strong diffusion planning baseline that achieves state-of-the-art results on standard offline RL benchmarks.

cs.LG↗

arXiv2025

Habitizing Diffusion Planning for Efficient and Effective Decision Making

Diffusion models have shown great promise in decision-making, also known as diffusion planning. However, the slow inference speeds limit their potential for broader real-world applications. Here, we introduce Habi, a general framework that transforms powerful but slow diffusion planning models into fast decision-making models, which mimics the cognitive process in the brain that costly goal-directed behavior gradually transitions to efficient habitual behavior with repetitive practice. Even using a laptop CPU, the habitized model can achieve an average 800+ Hz decision-making frequency (faster than previous diffusion planners by orders of magnitude) on standard offline reinforcement learning benchmarks D4RL, while maintaining comparable or even higher performance compared to its corresponding diffusion planner. Our work proposes a fresh perspective of leveraging powerful diffusion models for real-world decision-making tasks. We also provide robust evaluations and analysis, offering insights from both biological and engineering perspectives for efficient and effective decision-making.

cs.LG↗

arXiv2025

SimSort: A Data-Driven Framework for Spike Sorting by Large-Scale Electrophysiology Simulation

Spike sorting is an essential process in neural recording, which identifies and separates electrical signals from individual neurons recorded by electrodes in the brain, enabling researchers to study how specific neurons communicate and process information. Although there exist a number of spike sorting methods which have contributed to significant neuroscientific breakthroughs, many are heuristically designed, making it challenging to verify their correctness due to the difficulty of obtaining ground truth labels from real-world neural recordings. In this work, we explore a data-driven, deep learning-based approach. We begin by creating a large-scale dataset through electrophysiology simulations using biologically realistic computational models. We then present SimSort, a pretraining framework for spike sorting. Trained solely on simulated data, SimSort demonstrates zero-shot generalizability to real-world spike sorting tasks, yielding consistent improvements over existing methods across multiple benchmarks. These results highlight the potential of simulation-driven pretraining to enhance the robustness and scalability of spike sorting in experimental neuroscience.

q-bio.NC↗

arXiv2025

Toward Relative Positional Encoding in Spiking Transformers

Spiking neural networks (SNNs) are bio-inspired networks that mimic how neurons in the brain communicate through discrete spikes, which have great potential in various tasks due to their energy efficiency and temporal processing capabilities. SNNs with self-attention mechanisms (spiking Transformers) have recently shown great advancements in various tasks, and inspired by traditional Transformers, several studies have demonstrated that spiking absolute positional encoding can help capture sequential relationships for input data, enhancing the capabilities of spiking Transformers for tasks such as sequential modeling and image classification. However, how to incorporate relative positional information into SNNs remains a challenge. In this paper, we introduce several strategies to approximate relative positional encoding in spiking Transformers while preserving the binary nature of spikes. Firstly, we formally prove that encoding relative distances with Gray Code ensures that the binary representations of positional indices maintain a constant Hamming distance whenever their decimal values differ by a power of two, and we propose Gray-PE based on this property. In addition, we propose another RPE method called Log-PE, which combines the logarithmic form of the relative distance matrix directly into the spiking attention map. Furthermore, we extend our RPE methods to a two-dimensional form, making them suitable for processing image patches. We evaluate our RPE methods on various tasks, including time series forecasting, text classification, and patch-based image classification, and the experimental results demonstrate a satisfying performance gain by incorporating our RPE methods across many architectures. Our results provide fresh perspectives on designing spiking Transformers to advance their sequential modeling capability, thereby expanding their applicability across various domains.

cs.NE↗

arXiv2025

Mitigating Hallucinations in Large Vision-Language Models via DPO: On-Policy Data Hold the Key

Hallucination remains a major challenge for Large Vision-Language Models (LVLMs). Direct Preference Optimization (DPO) has gained increasing attention as a simple solution to hallucination issues. It directly learns from constructed preference pairs that reflect the severity of hallucinations in responses to the same prompt and image. Nonetheless, different data construction methods in existing works bring notable performance variations. We identify a crucial factor here: outcomes are largely contingent on whether the constructed data aligns on-policy w.r.t the initial (reference) policy of DPO. Theoretical analysis suggests that learning from off-policy data is impeded by the presence of KL-divergence between the updated policy and the reference policy. From the perspective of dataset distribution, we systematically summarize the inherent flaws in existing algorithms that employ DPO to address hallucination issues. To alleviate the problems, we propose On-Policy Alignment (OPA)-DPO framework, which uniquely leverages expert feedback to correct hallucinated responses and aligns both the original and expert-revised responses in an on-policy manner. Notably, with only 4.8k data, OPA-DPO achieves an additional reduction in the hallucination rate of LLaVA-1.5-7B: 13.26% on the AMBER benchmark and 5.39% on the Object-Hal benchmark, compared to the previous SOTA algorithm trained with 16k samples. Our implementation is available at https://github.com/zhyang2226/OPA-DPO.

cs.CV↗

arXiv2024

FG-SAT: Efficient Flow Graph for Encrypted Traffic Classification under Environment Shifts

Encrypted traffic classification plays a critical role in network security and management. Currently, mining deep patterns from side-channel contents and plaintext fields through neural networks is a major solution. However, existing methods have two major limitations: (1) They fail to recognize the critical link between transport layer mechanisms and applications, missing the opportunity to learn internal structure features for accurate traffic classification. (2) They assume network traffic in an unrealistically stable and singular environment, making it difficult to effectively classify real-world traffic under environment shifts. In this paper, we propose FG-SAT, the first end-to-end method for encrypted traffic analysis under environment shifts. We propose a key abstraction, the Flow Graph, to represent flow internal relationship structures and rich node attributes, which enables robust and generalized representation. Additionally, to address the problem of inconsistent data distribution under environment shifts, we introduce a novel feature selection algorithm based on Jensen-Shannon divergence (JSD) to select robust node attributes. Finally, we design a classifier, GraphSAT, which integrates GraphSAGE and GAT to deeply learn Flow Graph features, enabling accurate encrypted traffic identification. FG-SAT exhibits both efficient and robust classification performance under environment shifts and outperforms state-of-the-art methods in encrypted attack detection and application classification.

cs.CR↗

arXiv2024

The VEP Booster: A Closed-Loop AI System for Visual EEG Biomarker Auto-generation

Effective visual brain-machine interfaces (BMI) is based on reliable and stable EEG biomarkers. However, traditional adaptive filter-based approaches may suffer from individual variations in EEG signals, while deep neural network-based approaches may be hindered by the non-stationarity of EEG signals caused by biomarker attenuation and background oscillations. To address these challenges, we propose the Visual Evoked Potential Booster (VEP Booster), a novel closed-loop AI framework that generates reliable and stable EEG biomarkers under visual stimulation protocols. Our system leverages an image generator to refine stimulus images based on real-time feedback from human EEG signals, generating visual stimuli tailored to the preferences of primary visual cortex (V1) neurons and enabling effective targeting of neurons most responsive to stimuli. We validated our approach by implementing a system and employing steady-state visual evoked potential (SSVEP) visual protocols in five human subjects. Our results show significant enhancements in the reliability and utility of EEG biomarkers for all individuals, with the largest improvement in SSVEP response being 105%, the smallest being 28%, and the average increase being 76.5%. These promising results have implications for both clinical and technological applications

cs.CV↗

arXiv2024

Advancing Spiking Neural Networks for Sequential Modeling with Central Pattern Generators

Spiking neural networks (SNNs) represent a promising approach to developing artificial neural networks that are both energy-efficient and biologically plausible. However, applying SNNs to sequential tasks, such as text classification and time-series forecasting, has been hindered by the challenge of creating an effective and hardware-friendly spike-form positional encoding (PE) strategy. Drawing inspiration from the central pattern generators (CPGs) in the human brain, which produce rhythmic patterned outputs without requiring rhythmic inputs, we propose a novel PE technique for SNNs, termed CPG-PE. We demonstrate that the commonly used sinusoidal PE is mathematically a specific solution to the membrane potential dynamics of a particular CPG. Moreover, extensive experiments across various domains, including time-series forecasting, natural language processing, and image classification, show that SNNs with CPG-PE outperform their conventional counterparts. Additionally, we perform analysis experiments to elucidate the mechanism through which SNNs encode positional information and to explore the function of CPGs in the human brain. This investigation may offer valuable insights into the fundamental principles of neural computation. Our code is available at https://github.com/microsoft/SeqSNN.

cs.NE↗