Ali Shehper

I am a Research Scientist at Caltech, building AI systems to solve hard, research-level math problems.

Previously, I received my PhD in Physics from UT Austin, under the guidance of Jacques Distler and Andrew Neitkze.

AI Papers:

• What makes math problems hard for reinforcement learning: a case study,
  A. Shehper, A. M. Medina-Mardones, L. Fagan, B. Lewandowski, A. Gruen, Y. Qiu, P. Kucharski, Z. Wang, S. Gukov.
  Investigating reinforcement learning challenges in mathematical problem-solving.

Code:

• AC-Solver:
  A Python library for tackling long-horizon, ultra-sparse-reward RL environments, designed to accompany our case study.

• Sparse-Dictionary-Learning:
  An open-source implementation of Anthropic’s Towards Monosemanticity: Decomposing Language Models with Dictionary Learning.

• Neural Scaling Laws:
  An implementation of Scaling Laws for Neural Language Models, along with results from An Empirical Model of Large-Batch Training.

• Language Model Feature Browser:
  A visualizer for features learned by a 1-layer Language Model, with the GitHub Repository.

Physics / Math Papers:

• PhD Thesis: Aspects of Supersymmetric and Topological Quantum Field Theories.

• Nonabelian Twists of the D4 Theory:
  J. Distler, B. Ergun, A. Shehper (co-primary contributor, names in alphabetical order).

• Symmetries of 2d TQFTs and Equivariant Verlinde Formulae for General Groups:
  S. Gukov, D. Pei, C. Reid, A. Shehper (primary contributor, names in alphabetical order).

• Distinguishing 4d N=2 SCFTs:
  J. Distler, B. Ergun, A. Shehper (co-primary contributor, names in alphabetical order).

• Deformations of surface defect moduli spaces:
  A. Neitzke, A. Shehper (primary contributor, names in alphabetical order).

Contact and Collaboration

I am always open to discussions on reinforcement learning, AI interpretability, and related areas of physics and mathematics. Feel free to reach out for potential collaborations, research opportunities, or simply to connect.

Google Scholar / GitHub / Twitter / LinkedIn