Working on video understanding project by involving causal representation learning and identifiability theory. The purpose of this project is to provide a more principled, causal pipeline for video-based computer vision tasks.

Research in topics e.g. Machine Learning, Causality, Natural Language Processing and Optimization etc. Full scholarship support from the university.GPA: 3.95/4.Publication: Unsupervised Sampling Promoting for Stochastic Human Trajectory Prediction, CVPR 2023, Co-first author.

I was one of the main contributors of an open source repository developed mainly with MindSpore. (https://gitee.com/mindspore/models/tree/master/research/nlp/gpt2) This repo reproduced OpenAI's GPT-2 from scratch and the model achieved the same performance as the original model in translation, language modeling and summarization tasks. Specifically, I took in charge of text generation framework and deployment tools. I also fine tuned this GPT-2 model in summarization task on NVIDIA Docker and HUAWEI Ascend devices. Moreover, I devoted in model compression and other techniques which could help accelerate inferring on device like mobile phones and edge computing devices. What's more, I engaged in applying algorithms which are originated from quantum physics and quantum information (e.g. tensor network) on machine learning applications on edge computing device (e.g. ARM based raspberry pi).

TA of Natural Language Understading. Gave a lecture about neural machine translation. Also gave a talk about pretrained language model and its application. Help design assignment and grading.