Ali Shehper


View My GitHub Profile

Ali Shehper

I am a postdoc in Physics at Rutgers University, where my research focuses on using Reinforcement Learning and Natural Langauge Processing to gain new insights into Math problems. For example, with my collaborators, I am studying the use of Policy Gradient algorithms to solve potential counterexamples of the Andrews-Curtis conjecture.

I am also interested in studying the internals of deep learning models. I am currently implementing Anthropic’s paper on the use of Sparse AutoEncoders to discover monosemantic features in language models. You may check out my GitHub repository here.

I also recently implemented the scaling laws of Kaplan-McCandlish et al for large language models. The repository for this project is here.

Prior to my time at Rutgers, I studied theoretical physics at UT Austin, where my Ph.D. focused on theoretical aspects of quantum field theories. A list of my research papers can be found here.

Please feel free to reach out if you are also interested in similar things or just want to chat.

Google Scholar / GitHub / Twitter / LinkedIn