Zilong Zheng

MathBench assesses LLMs' math capabilities across various educational stages.

DiveR-CT improves automated red teaming for better safety assessments.

A novel approach enhances Transformer models for better long text processing.

New benchmark assesses how video-language models handle inaccuracies effectively.

A new method helps robots navigate and orient correctly for tasks.

This method enhances visual reasoning by implementing verification at each reasoning step.

A framework using memory tokens improves video understanding and interaction.