Wengang Zhou

4 papers University of Science and Technology of China (USTC)

• Enhancing the Outcome Reward-based RL Training of MLLMs with Self-Consistency Sampling
• Image as a World: Generating Interactive World from Single Image via Panoramic Video Generation
• Mitigating Hallucination in VideoLLMs via Temporal-Aware Activation Engineering
• Optimizing Distributional Geometry Alignment with Optimal Transport for Generative Dataset Distillation