Chih-Hao Fang

Chih-Hao Fang

Post Doc in Computer Science at Purdue University.

About Me

** I am currently open to opportunities in the roles of an Applied Scientist or Machine Learning Researcher. I welcome any interested parties to connect with me on LinkedIn or via my email at chihhaofang19@gmail.com. **

Possessing a rich background in Machine Learning research, I bring to the table comprehensive expertise in Convex, Non-Convex, and Constrained Optimization, Causal Inference, and Deep Learning. My technical acumen shines in my ability to design and implement Convex and Non-Convex solvers which streamline the training process of machine learning models. In this capacity, I have accumulated vast experience with open-source platforms like Pytorch and Tensorflow, allowing me to modify and even extend their functionalities.

In the healthcare industry, my work has been pivotal in utilizing constrained optimization solvers for disease phenotyping. This work has paved the way for an innovative interpretable treatment recommendation model.

I’ve also made significant contributions in the development of a causal reinforcement learning algorithm. This unique algorithm leverages causal inference principles to forecast more accurate potential treatment outcomes.

My involvement in Deep Learning has resulted in the creation of a model that outperforms the current state-of-the-art models in predicting regulatory elements.

Recently, I’ve further refined my optimization skills by creating a constrained optimization solver. This solver can identify pivotal features corresponding to multidimensional feature and outcome spaces. Recognized by experts in the field, this represents a high-dimensional extension of L1 lasso regression, thereby enhancing the model’s performance and explainability significantly.

At present, my focus is on designing cutting-edge deep learning models to solve intricate Natural Language Processing problems. A notable accomplishment is the unique framework I’ve developed for news framing prediction. This framework adopts a graph-based approach that converts sentences into graph nodes, utilizes transformer models for embeddings, and extracts relationships from the ASER Knowledge Graph. A trained Graph Neural Network is then deployed to fine-tune transformers and accurately predict the framing of news articles.

In addition to my research work, I serve as a conference reviewer for highly respected forums such as NeurIPS and ICML, contributing to the shared body of knowledge in this rapidly progressing field. In my upcoming role, I aim to utilize my specialized expertise to continue propelling the boundaries of research forward.

News