Chenglong Chu

Hello 🙋‍♂️！

I am Chenglong Chu, a Master student at INCODAT Lab, Dalian University of Technology, advised by Prof. Fangming Zhong. I received my B.Eng. in Software Engineering from Changzhou University.

My current research interests focus on multi-modal feature alignment and multi-modal large model pre-training. With the increasingly strong trend of multi-modal integration in the community, I hope to focus on the following areas in the future:

• Lightweight and efficient generative large models: Most of the current mainstream large models are bloated and clumsy, and the Sora model requires more than one hour of computation time to generate one minute of video, and the effect is not stable. By studying the compression acceleration methods for generative large models, we can reduce the deployment cost of large models and make large models better used in the real world. Furthermore, the recent excellent performance of llama3 8B demonstrates the potential of small models to have the same representation power as large models.

• Multi-modality large models & Embodied Intelligence: Traditional computer vision research paradigms struggle to adapt flexibly to the complex physical rules of the real world. Large models based on multi-modality are poised to break through these limitations in the future. The outcomes of such research hold significant transformational value and promise broad application possibilities.

News

• [Feb. 2024] Code and model weights of HEH are released!
• [Nov. 2023] Received 2023 Intel Scholarship
• [Oct. 2023] Received Individual award for postgraduate research innovation
• [Jun. 2023] One paper about cross-modal retrival is accepted to ACM MM 2023.

Publications

© Chenglong Chu, 2024