arXiv subjects

Wei-Long Zheng

Publications and source records attributed to Wei-Long Zheng.

18 recordsLinked to original sources

arXiv2026

Visualizing the Invisible: Generative Visual Grounding Empowers Universal EEG Understanding in MLLMs

Leveraging the universal representations of pre-trained LLMs and MLLMs offers a promising path toward brain foundation models. However, visually-evoked EEG datasets remain scarce, leading existing methods to align neural signals mainly with abstract text, a lossy translation that may discard fine-grained perceptual information encoded in brain activity. We propose Generative Visual Grounding (GVG), a framework that visualizes the invisible by using an EEG-to-image generative model as a visual translator. Instead of forcing EEG into text alone, GVG hallucinates instance-specific proxy images for non-visual EEG, providing structured visual contexts that allow MLLMs to exploit their visual priors for clinical-state interpretation. We validate this idea on two MLLM backbones, GVG-X-Omni and GVG-Janus. Image-only alignment is already competitive: the lightweight GVG-X-Omni matches 1.7B-parameter text-aligned baselines while tuning only 170M parameters on a frozen 7B backbone. We further extend GVG-Janus with trimodal Image+Text alignment, where text supplies categorical semantic anchors and visual proxies enrich neural representations with perceptual details. Experiments show consistent gains in EEG understanding and visual generation, suggesting visual proxy grounding as an effective complement to textual alignment.

cs.AI↗

arXiv2026

Persistent Visual Memory: Sustaining Perception for Deep Generation in LVLMs

While autoregressive Large Vision-Language Models (LVLMs) demonstrate remarkable proficiency in multimodal tasks, they face a "Visual Signal Dilution" phenomenon, where the accumulation of textual history expands the attention partition function, causing visual attention to decay inversely with generated sequence length. To counteract this, we propose Persistent Visual Memory (PVM), a lightweight learnable module designed to strengthen sustained, on-demand access to visual evidence. Integrated as a parallel branch alongside the Feed-Forward Network (FFN) in LVLMs, PVM establishes a distance-agnostic retrieval pathway that directly provides visual embeddings for enhanced visual perception, thereby structurally mitigating the signal suppression inherent to deep generation. Extensive experiments on Qwen3-VL models demonstrate that PVM brings notable improvements with negligible parameter overhead, delivering consistent average accuracy gains across both 4B and 8B scales, particularly in complex reasoning tasks that demand persistent visual perception. Furthermore, in-depth analysis reveals that PVM shows improved robustness in longer generations and accelerates internal prediction convergence.

cs.CV↗

arXiv2026

Measuring Audio's Impact on Correctness: Audio-Contribution-Aware Post-Training of Large Audio Language Models

Large Audio Language Models (LALMs) represent an important frontier in multimodal AI, addressing diverse audio tasks. Recently, post-training of LALMs has received increasing attention due to significant performance improvements over foundation models. While single-stage post-training such as reinforcement learning (RL) has demonstrated promising results, multi-stage approaches such as supervised fine-tuning (SFT) followed by RL remain suboptimal. The allocation of data across multiple training stages to maximize LALM capabilities has not been fully explored, and large-scale, high-quality datasets for such research are also lacking. To address these problems, we firstly present AudioMCQ, a comprehensive audio multiple-choice question dataset comprising 571k samples with two kinds of chain-of-thought annotations. Secondly, we investigate the prevalent zero audio-contribution phenomenon in LALMs, where models derive correct answers solely from textual information without processing audio content. We propose Audio-Contribution Filtering to partition data into weak and strong audio-contribution subsets. Based on these insights, we develop two effective post-training paradigms: Weak-to-Strong (SFT on weak audio-contribution data followed by RL on strong audio-contribution data) and Mixed-to-Strong (SFT on mixed audio-contribution data followed by RL on strong audio-contribution data). We achieve first place in the DCASE 2025 Audio-Question-Answering challenge by using AudioMCQ. Additionally, leveraging our dataset with different training strategies, we achieve 78.2\% on MMAU-test-mini, 75.6\% on MMAU, 67.1\% on MMAR, and 70.7\% on MMSU, establishing new state-of-the-art performance.

eess.AS↗

arXiv2026

MindCine: Multimodal EEG-to-Video Reconstruction with Large-Scale Pretrained Models

Reconstructing human dynamic visual perception from electroencephalography (EEG) signals is of great research significance since EEG's non-invasiveness and high temporal resolution. However, EEG-to-video reconstruction remains challenging due to: 1) Single Modality: existing studies solely align EEG signals with the text modality, which ignores other modalities and are prone to suffer from overfitting problems; 2) Data Scarcity: current methods often have difficulty training to converge with limited EEG-video data. To solve the above problems, we propose a novel framework MindCine to achieve high-fidelity video reconstructions on limited data. We employ a multimodal joint learning strategy to incorporate beyond-text modalities in the training stage and leverage a pre-trained large EEG model to relieve the data scarcity issue for decoding semantic information, while a Seq2Seq model with causal attention is specifically designed for decoding perceptual information. Extensive experiments demonstrate that our model outperforms state-of-the-art methods both qualitatively and quantitatively. Additionally, the results underscore the effectiveness of the complementary strengths of different modalities and demonstrate that leveraging a large-scale EEG model can further enhance reconstruction performance by alleviating the challenges associated with limited data.

cs.CV↗

arXiv2025

MindCross: Fast New Subject Adaptation with Limited Data for Cross-subject Video Reconstruction from Brain Signals

Reconstructing video from brain signals is an important brain decoding task. Existing brain decoding frameworks are primarily built on a subject-dependent paradigm, which requires large amounts of brain data for each subject. However, the expensive cost of collecting brain-video data causes severe data scarcity. Although some cross-subject methods being introduced, they often overfocus with subject-invariant information while neglecting subject-specific information, resulting in slow fine-tune-based adaptation strategy. To achieve fast and data-efficient new subject adaptation, we propose MindCross, a novel cross-subject framework. MindCross's N specific encoders and one shared encoder are designed to extract subject-specific and subject-invariant information, respectively. Additionally, a Top-K collaboration module is adopted to enhance new subject decoding with the knowledge learned from previous subjects' encoders. Extensive experiments on fMRI/EEG-to-video benchmarks demonstrate MindCross's efficacy and efficiency of cross-subject decoding and new subject adaptation using only one model.

cs.MM↗

arXiv2025

mixEEG: Enhancing EEG Federated Learning for Cross-subject EEG Classification with Tailored mixup

The cross-subject electroencephalography (EEG) classification exhibits great challenges due to the diversity of cognitive processes and physiological structures between different subjects. Modern EEG models are based on neural networks, demanding a large amount of data to achieve high performance and generalizability. However, privacy concerns associated with EEG pose significant limitations to data sharing between different hospitals and institutions, resulting in the lack of large dataset for most EEG tasks. Federated learning (FL) enables multiple decentralized clients to collaboratively train a global model without direct communication of raw data, thus preserving privacy. For the first time, we investigate the cross-subject EEG classification in the FL setting. In this paper, we propose a simple yet effective framework termed mixEEG. Specifically, we tailor the vanilla mixup considering the unique properties of the EEG modality. mixEEG shares the unlabeled averaged data of the unseen subject rather than simply sharing raw data under the domain adaptation setting, thus better preserving privacy and offering an averaged label as pseudo-label. Extensive experiments are conducted on an epilepsy detection and an emotion recognition dataset. The experimental result demonstrates that our mixEEG enhances the transferability of global model for cross-subject EEG classification consistently across different datasets and model architectures. Code is published at: https://github.com/XuanhaoLiu/mixEEG.

eess.SP↗

arXiv2024

Combining Induction and Transduction for Abstract Reasoning

When learning an input-output mapping from very few examples, is it better to first infer a latent function that explains the examples, or is it better to directly predict new test outputs, e.g. using a neural network? We study this question on ARC by training neural models for induction (inferring latent functions) and transduction (directly predicting the test output for a given test input). We train on synthetically generated variations of Python programs that solve ARC training tasks. We find inductive and transductive models solve different kinds of test problems, despite having the same training problems and sharing the same neural architecture: Inductive program synthesis excels at precise computations, and at composing multiple concepts, while transduction succeeds on fuzzier perceptual concepts. Ensembling them approaches human-level performance on ARC.

cs.LG↗

arXiv2024

FuseAnyPart: Diffusion-Driven Facial Parts Swapping via Multiple Reference Images

Facial parts swapping aims to selectively transfer regions of interest from the source image onto the target image while maintaining the rest of the target image unchanged. Most studies on face swapping designed specifically for full-face swapping, are either unable or significantly limited when it comes to swapping individual facial parts, which hinders fine-grained and customized character designs. However, designing such an approach specifically for facial parts swapping is challenged by a reasonable multiple reference feature fusion, which needs to be both efficient and effective. To overcome this challenge, FuseAnyPart is proposed to facilitate the seamless "fuse-any-part" customization of the face. In FuseAnyPart, facial parts from different people are assembled into a complete face in latent space within the Mask-based Fusion Module. Subsequently, the consolidated feature is dispatched to the Addition-based Injection Module for fusion within the UNet of the diffusion model to create novel characters. Extensive experiments qualitatively and quantitatively validate the superiority and robustness of FuseAnyPart. Source codes are available at https://github.com/Thomas-wyh/FuseAnyPart.

cs.CV↗

arXiv2024

Code Repair with LLMs gives an Exploration-Exploitation Tradeoff

Iteratively improving and repairing source code with large language models (LLMs), known as refinement, has emerged as a popular way of generating programs that would be too complex to construct in one shot. Given a bank of test cases, together with a candidate program, an LLM can improve that program by being prompted with failed test cases. But it remains an open question how to best iteratively refine code, with prior work employing simple greedy or breadth-first strategies. We show here that refinement exposes an explore-exploit tradeoff: exploit by refining the program that passes the most test cases, or explore by refining a lesser considered program. We frame this as an arm-acquiring bandit problem, which we solve with Thompson Sampling. The resulting LLM-based program synthesis algorithm is broadly applicable: Across loop invariant synthesis, visual reasoning puzzles, and competition programming problems, we find that our new method can solve more problems using fewer language model calls.

cs.SE↗

arXiv2024

Professor X: Manipulating EEG BCI with Invisible and Robust Backdoor Attack

While electroencephalogram (EEG) based brain-computer interface (BCI) has been widely used for medical diagnosis, health care, and device control, the safety of EEG BCI has long been neglected. In this paper, we propose Professor X, an invisible and robust "mind-controller" that can arbitrarily manipulate the outputs of EEG BCI through backdoor attack, to alert the EEG community of the potential hazard. However, existing EEG attacks mainly focus on single-target class attacks, and they either require engaging the training stage of the target BCI, or fail to maintain high stealthiness. Addressing these limitations, Professor X exploits a three-stage clean label poisoning attack: 1) selecting one trigger for each class; 2) learning optimal injecting EEG electrodes and frequencies strategy with reinforcement learning for each trigger; 3) generating poisoned samples by injecting the corresponding trigger's frequencies into poisoned data for each class by linearly interpolating the spectral amplitude of both data according to previously learned strategies. Experiments on datasets of three common EEG tasks demonstrate the effectiveness and robustness of Professor X, which also easily bypasses existing backdoor defenses.

cs.CR↗

arXiv2024

Focused State Recognition Using EEG with Eye Movement-Assisted Annotation

With the rapid advancement in machine learning, the recognition and analysis of brain activity based on EEG and eye movement signals have attained a high level of sophistication. Utilizing deep learning models for learning EEG and eye movement features proves effective in classifying brain activities. A focused state indicates intense concentration on a task or thought. Distinguishing focused and unfocused states can be achieved through eye movement behaviors, reflecting variations in brain activities. By calculating binocular focusing point disparity in eye movement signals and integrating relevant EEG features, we propose an annotation method for focused states. The resulting comprehensive dataset, derived from raw data processed through a bio-acquisition device, includes both EEG features and focused labels annotated by eye movements. Extensive training and testing on several deep learning models, particularly the Transformer, yielded a 90.16% accuracy on the subject-dependent experiments. The validity of this approach was demonstrated, with cross-subject experiments, key frequency band and brain region analyses confirming its generalizability and providing physiological explanations.

cs.HC↗

arXiv2023

Seeing through the Brain: Image Reconstruction of Visual Perception from Human Brain Signals

Seeing is believing, however, the underlying mechanism of how human visual perceptions are intertwined with our cognitions is still a mystery. Thanks to the recent advances in both neuroscience and artificial intelligence, we have been able to record the visually evoked brain activities and mimic the visual perception ability through computational approaches. In this paper, we pay attention to visual stimuli reconstruction by reconstructing the observed images based on portably accessible brain signals, i.e., electroencephalography (EEG) data. Since EEG signals are dynamic in the time-series format and are notorious to be noisy, processing and extracting useful information requires more dedicated efforts; In this paper, we propose a comprehensive pipeline, named NeuroImagen, for reconstructing visual stimuli images from EEG signals. Specifically, we incorporate a novel multi-level perceptual information decoding to draw multi-grained outputs from the given EEG data. A latent diffusion model will then leverage the extracted information to reconstruct the high-resolution visual stimuli images. The experimental results have illustrated the effectiveness of image reconstruction and superior quantitative performance of our proposed method.

eess.IV↗

arXiv2020

Investigating EEG-Based Functional Connectivity Patterns for Multimodal Emotion Recognition

Compared with the rich studies on the motor brain-computer interface (BCI), the recently emerging affective BCI presents distinct challenges since the brain functional connectivity networks involving emotion are not well investigated. Previous studies on emotion recognition based on electroencephalography (EEG) signals mainly rely on single-channel-based feature extraction methods. In this paper, we propose a novel emotion-relevant critical subnetwork selection algorithm and investigate three EEG functional connectivity network features: strength, clustering coefficient, and eigenvector centrality. The discrimination ability of the EEG connectivity features in emotion recognition is evaluated on three public emotion EEG datasets: SEED, SEED-V, and DEAP. The strength feature achieves the best classification performance and outperforms the state-of-the-art differential entropy feature based on single-channel analysis. The experimental results reveal that distinct functional connectivity patterns are exhibited for the five emotions of disgust, fear, sadness, happiness, and neutrality. Furthermore, we construct a multimodal emotion recognition model by combining the functional connectivity features from EEG and the features from eye movements or physiological signals using deep canonical correlation analysis. The classification accuracies of multimodal emotion recognition are 95.08/6.42% on the SEED dataset, 84.51/5.11% on the SEED-V dataset, and 85.34/2.90% and 86.61/3.76% for arousal and valence on the DEAP dataset, respectively. The results demonstrate the complementary representation properties of the EEG connectivity features with eye movement data. In addition, we find that the brain networks constructed with 18 channels achieve comparable performance with that of the 62-channel network in multimodal emotion recognition and enable easier setups for BCI systems in real scenarios.

cs.HC↗

arXiv2019

Multimodal Emotion Recognition Using Deep Canonical Correlation Analysis

Multimodal signals are more powerful than unimodal data for emotion recognition since they can represent emotions more comprehensively. In this paper, we introduce deep canonical correlation analysis (DCCA) to multimodal emotion recognition. The basic idea behind DCCA is to transform each modality separately and coordinate different modalities into a hyperspace by using specified canonical correlation analysis constraints. We evaluate the performance of DCCA on five multimodal datasets: the SEED, SEED-IV, SEED-V, DEAP, and DREAMER datasets. Our experimental results demonstrate that DCCA achieves state-of-the-art recognition accuracy rates on all five datasets: 94.58% on the SEED dataset, 87.45% on the SEED-IV dataset, 84.33% and 85.62% for two binary classification tasks and 88.51% for a four-category classification task on the DEAP dataset, 83.08% on the SEED-V dataset, and 88.99%, 90.57%, and 90.67% for three binary classification tasks on the DREAMER dataset. We also compare the noise robustness of DCCA with that of existing methods when adding various amounts of noise to the SEED-V dataset. The experimental results indicate that DCCA has greater robustness. By visualizing feature distributions with t-SNE and calculating the mutual information between different modalities before and after using DCCA, we find that the features transformed by DCCA from different modalities are more homogeneous and discriminative across emotions.

cs.LG↗

arXiv2018

Semi-supervised Deep Generative Modelling of Incomplete Multi-Modality Emotional Data

There are threefold challenges in emotion recognition. First, it is difficult to recognize human's emotional states only considering a single modality. Second, it is expensive to manually annotate the emotional data. Third, emotional data often suffers from missing modalities due to unforeseeable sensor malfunction or configuration issues. In this paper, we address all these problems under a novel multi-view deep generative framework. Specifically, we propose to model the statistical relationships of multi-modality emotional data using multiple modality-specific generative networks with a shared latent space. By imposing a Gaussian mixture assumption on the posterior approximation of the shared latent variables, our framework can learn the joint deep representation from multiple modalities and evaluate the importance of each modality simultaneously. To solve the labeled-data-scarcity problem, we extend our multi-view model to semi-supervised learning scenario by casting the semi-supervised classification problem as a specialized missing data imputation task. To address the missing-modality problem, we further extend our semi-supervised multi-view model to deal with incomplete data, where a missing view is treated as a latent variable and integrated out during inference. This way, the proposed overall framework can utilize all available (both labeled and unlabeled, as well as both complete and incomplete) data to improve its generalization ability. The experiments conducted on two real multi-modal emotion datasets demonstrated the superiority of our framework.

eess.SP↗

arXiv2016

A Multimodal Approach to Estimating Vigilance Using EEG and Forehead EOG

Objective. Covert aspects of ongoing user mental states provide key context information for user-aware human computer interactions. In this paper, we focus on the problem of estimating the vigilance of users using EEG and EOG signals. Approach. To improve the feasibility and wearability of vigilance estimation devices for real-world applications, we adopt a novel electrode placement for forehead EOG and extract various eye movement features, which contain the principal information of traditional EOG. We explore the effects of EEG from different brain areas and combine EEG and forehead EOG to leverage their complementary characteristics for vigilance estimation. Considering that the vigilance of users is a dynamic changing process because the intrinsic mental states of users involve temporal evolution, we introduce continuous conditional neural field and continuous conditional random field models to capture dynamic temporal dependency. Main results. We propose a multimodal approach to estimating vigilance by combining EEG and forehead EOG and incorporating the temporal dependency of vigilance into model training. The experimental results demonstrate that modality fusion can improve the performance compared with a single modality, EOG and EEG contain complementary information for vigilance estimation, and the temporal dependency-based models can enhance the performance of vigilance estimation. From the experimental results, we observe that theta and alpha frequency activities are increased, while gamma frequency activities are decreased in drowsy states in contrast to awake states. Significance. The forehead setup allows for the simultaneous collection of EEG and EOG and achieves comparative performance using only four shared electrodes in comparison with the temporal and posterior sites.

cs.HC↗

arXiv2016

Multimodal Emotion Recognition Using Multimodal Deep Learning

To enhance the performance of affective models and reduce the cost of acquiring physiological signals for real-world applications, we adopt multimodal deep learning approach to construct affective models from multiple physiological signals. For unimodal enhancement task, we indicate that the best recognition accuracy of 82.11% on SEED dataset is achieved with shared representations generated by Deep AutoEncoder (DAE) model. For multimodal facilitation tasks, we demonstrate that the Bimodal Deep AutoEncoder (BDAE) achieves the mean accuracies of 91.01% and 83.25% on SEED and DEAP datasets, respectively, which are much superior to the state-of-the-art approaches. For cross-modal learning task, our experimental results demonstrate that the mean accuracy of 66.34% is achieved on SEED dataset through shared representations generated by EEG-based DAE as training samples and shared representations generated by eye-based DAE as testing sample, and vice versa.

cs.HC↗

arXiv2016

Identifying Stable Patterns over Time for Emotion Recognition from EEG

In this paper, we investigate stable patterns of electroencephalogram (EEG) over time for emotion recognition using a machine learning approach. Up to now, various findings of activated patterns associated with different emotions have been reported. However, their stability over time has not been fully investigated yet. In this paper, we focus on identifying EEG stability in emotion recognition. To validate the efficiency of the machine learning algorithms used in this study, we systematically evaluate the performance of various popular feature extraction, feature selection, feature smoothing and pattern classification methods with the DEAP dataset and a newly developed dataset for this study. The experimental results indicate that stable patterns exhibit consistency across sessions; the lateral temporal areas activate more for positive emotion than negative one in beta and gamma bands; the neural patterns of neutral emotion have higher alpha responses at parietal and occipital sites; and for negative emotion, the neural patterns have significant higher delta responses at parietal and occipital sites and higher gamma responses at prefrontal sites. The performance of our emotion recognition system shows that the neural patterns are relatively stable within and between sessions.

cs.HC↗