About Me

I am a second-year graduate student majored in data science at USC. I am very interest is interpretable ML from the perspective of mining the causal structure of the model and that could help in generalizing the model to out-of-distribution commonly in the setting of lifelong learning, zero and few-shot learning. Also, I am passionate about how we could utilize the causal discovery models in natural science, especially biology, to make new advances.

Education

University of Southern California 2017-2021 CS, Bachelor of Science
University of Southern California 2021-2023 DS, Master of Science

Research Experience

Invariant Structure Learning for Better Generalization and Causal Explainability
Yunhao Ge, Sercan O Arik, Jinsung Yoon, Ao Xu, Laurent Itti, Tomas Pfister
Devoted to developing the Invariant Structure Learning model capable of mining causal relationships on observatioanl tabular data that is based on the combining the constrained inviriant risk minimization (Arjovsky, et) and the directed acyclic graph constrains (Zheng ,et)

Shared Knowledge Lifelong Learning (just submmited to ICLR)
Yunhao Ge, Yuecheng Li, Di Wu, Ao Xu, Adam M Jones, Amanda Sofie Rios, Iordanis Fostiropoulos, shixian wen, Po-Hsuan Huang, Zachary William Murdock, Kiran Lekkala, Gozde Sahin, Sumedh Anand Sontakke, Laurent Itti
A total of 102 image classification tasks for the lifelong learning setting and a novel appraoch combining a GMMC task mapper with benificial biasese that are both efficienct and effective in the prediction accuracy on the 102 tasks compared with other 18 methods (the reularization-based, the rehersal-based and the parameter expanding approaches)

Research Assistant at USC Rong Lu Lab May. 2022 – Aug.2022 Conducted cell segmentation and quantification on microscopy images of bone marrow sections with U-net and cell Profiler.
Trajectory inference for hematopoiesis process by analyzing the scRNA data on HSCs.
Applied random forest, partial linear regression, and principal component regression to identity a pattern of HSCs differential process.

Projects

Recommendation System: H&M Personalized Fashion Recommendation Marh. 2022-May.2022
Built an item-based collaborative filtering system combined with clustering to make personalized recommendation for H&M customers and won a bronze for the Kaggle H&M personalized fashion recommendation.
Built a similar hybrid recommendation system for Yelp Restaurant review prediction by item-based approach and content-based modeling with sentiment analysis and feature engineering. The model achieved an RMSE < 1 on test set.

USC Directed Research program instructed by Iordanis Fostriopoulos Aug. 2021 – Dec.2021
Worked on a novel Depthwise Quantization model based on a hierarchical VQ-VAE for generative tasks, in this case, achieving a lossless compression of the image data by learning the residuals of its reconstructions from various lossy compression algorithms (e.g., BGP, VQ-VAE)

Intelligent small-Go Player: Sep. 2021 – Oct. 2021 Implemented the Q-learning algorithm to learn the utilities of moves under different configurations of a 5*5 board by playing against a random player, a greedy player, and a minimax player. After learning and fine-tuning of the learning rate, the Q-learner achieved over 90%-win rate against the minimax players with search-depth of 3 on a 5 by 5 board.