Shuhui Wang

Full Professor@Institute of Computing Technology, Chinese Academy of Sciences·中国·OpenReview

研究方向

Dynamics-inspired Neural Computing · Graph Neural Networks · multimodal knowledge base · Generative Adversarial Nets · Weakly-supervised learning · image/video captioning · Visual question answering · Domain adaptation · Vision-Language · Image/video understanding · Multi-modal fusion and analysis

Shuhui Wang

Test-Time Scaling via Metric Geometry for LLM Reasoning

Adaptive Nonlinear Compression for Large Foundation Models

DyG$^2$T: Modeling Object Dynamics with 3D Gaussian Temporal-Spatial Particle Graph Transformer

TR-MERGING: TRAINING-FREE ROUTER FOR MODEL MERGING

Pre-training Pure GNNs as Graph Learners

MIN-MERGING: MERGE THE IMPORTANT NEURONS FOR MODEL MERGING

Enhancing Pre-trained Representation Classifiability can Boost its Interpretability

Edit Less, Achieve More: Dynamic Sparse Neuron Masking for Lifelong Knowledge Editing in LLMs

Masked Temporal Interpolation Diffusion for Procedure Planning in Instructional Videos

VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept Set

Relieving the Over-Aggregating Effect in Graph Transformers

Learning Fine-Grained Representations through Textual Token Disentanglement in Composed Video Retrieval

Kolmogorov-Arnold Networks with Variable Function Basis