Author name cluster

Wen Luo

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

1 author row

AAAI Conference 2026 Conference Paper

ViTCoP: Accelerating Large Vision-Language Models via Visual and Textual Semantic Collaborative Pruning

Wen Luo
Peng Chen
Xiaotao Huang
LiQun Huang

Large Vision-Language Models (LVLMs) incur high computational costs due to significant redundancy in their visual tokens. To effectively reduce this cost, researchers have proposed various visual token pruning methods. However, existing methods are generally limited, either losing critical visual information prematurely due to pruning in the vision encoder, or leading to information redundancy among the selected tokens due to pruning in the Large Language Models (LLMs). To address these challenges, we propose a Visual and Textual Semantic Collaborative Pruning framework (ViTCoP) that combines redundancy filtering in the vision encoder with step-wise co-pruning within the LLM based on its hierarchical characteristics, to efficiently preserve critical and informationally diverse visual tokens. Meanwhile, to ensure compatibility with acceleration techniques like FlashAttention, we introduce the L2 norm of K-vectors as the token saliency metric in the LLM. Extensive experiments on various Large Vision-Language Models demonstrate that ViTCoP not only achieves state-of-the-art performance surpassing existing methods on both image and video understanding tasks, but also significantly reduces model inference latency and GPU memory consumption. Notably, its performance advantage over other methods becomes even more pronounced under extreme pruning rates.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

TimE: A Multi-level Benchmark for Temporal Reasoning of LLMs in Real-World Scenarios

Shaohang Wei
Wei Li
Feifan Song
Wen Luo
Tianyi Zhuang
Haochen Tan
Zhijiang Guo
Houfeng Wang

Temporal reasoning is pivotal for Large Language Models (LLMs) to comprehend the real world. However, existing works neglect the real-world challenges for temporal reasoning: (1) intensive temporal information, (2) fast-changing event dynamics, and (3) complex temporal dependencies in social interactions. To bridge this gap, we propose a multi-level benchmark TimE, designed for temporal reasoning in real-world scenarios. TimE consists of 38, 522 QA pairs, covering 3 levels with 11 fine-grained sub-tasks. This benchmark encompasses 3 sub-datasets reflecting different real-world challenges: TimE-Wiki, TimE-News, and TimE-Dial. We conduct extensive experiments on reasoning models and non-reasoning models. And we conducted an in-depth analysis of temporal reasoning performance across diverse real-world scenarios and tasks, and summarized the impact of test-time scaling on temporal reasoning capabilities. Additionally, we release TimE-Lite, a human-annotated subset to foster future research and standardized evaluation in temporal reasoning.

PDF Details

YNIMG Journal 2024 Journal Article

Precise detection of awareness in disorders of consciousness using deep learning framework

Huan Yang
Hang Wu
Lingcong Kong
Wen Luo
Qiuyou Xie
Jiahui Pan
Wuxiu Quan
Lianting Hu

Diagnosis of disorders of consciousness (DOC) remains a formidable challenge. Deep learning methods have been widely applied in general neurological and psychiatry disorders, while limited in DOC domain. Considering the successful use of resting-state functional MRI (rs-fMRI) for evaluating patients with DOC, this study seeks to explore the conjunction of deep learning techniques and rs-fMRI in precisely detecting awareness in DOC. We initiated our research with a benchmark dataset comprising 140 participants, including 76 unresponsive wakefulness syndrome (UWS), 25 minimally conscious state (MCS), and 39 Controls, from three independent sites. We developed a cascade 3D EfficientNet-B3-based deep learning framework tailored for discriminating MCS from UWS patients, referred to as "DeepDOC", and compared its performance against five state-of-the-art machine learning models. We also included an independent dataset consists of 11 DOC patients to test whether our model could identify patients with cognitive motor dissociation (CMD), in which DOC patients were behaviorally diagnosed unconscious but could be detected conscious by brain computer interface (BCI) method. Our results demonstrate that DeepDOC outperforms the five machine learning models, achieving an area under curve (AUC) value of 0.927 and accuracy of 0.861 for distinguishing MCS from UWS patients. More importantly, DeepDOC excels in CMD identification, achieving an AUC of 1 and accuracy of 0.909. Using gradient-weighted class activation mapping algorithm, we found that the posterior cortex, encompassing the visual cortex, posterior middle temporal gyrus, posterior cingulate cortex, precuneus, and cerebellum, as making a more substantial contribution to classification compared to other brain regions. This research offers a convenient and accurate method for detecting covert awareness in patients with MCS and CMD using rs-fMRI data.

Details DOI