Espen Slettnes

MIT, Class of 2027

I research mathematics, computer science, and artificial intelligence. I expect to graduate in May 2027 with an M.Eng. in Electrical Engineering and Computer Science, a B.S. inM athematics, and a B.S. in Artificial Intelligence and Decision-Making.

My research interests center on applications of mathematical ideas to computational problems. I am particularly interested in machine learning, reinforcement learning, transformer interpretability, computer graphics and vision, and numerical methods.

Some of the questions I am currently exploring include:

• How differently do different RoPE frequencies within an attention head behave? How does the tokenization scheme affect this behavior?
• Is reward enough?
• When will training an LLM on specific tasks improve its performance on unrelated tasks?
• Can a neural network learn to be solve Sokoban through reinforcement learning by putting it in competition against another network trained to generate puzzles? Will the generated puzzles be diverse, interesting, or difficult for humans?
• What can be inferred about market dynamics from the options market?
• How can we compute numerical solutions to elliptic PDEs more efficiently when the equations have complex boundary conditions?
• What makes an API designed to be used in an embedded scripting language effective?
• Is Sokoban decidable when the player controls two characters simultaneously with the same inputs, but one character moves with a one-second delay?
• Can perfect harmonies be achieved, even in the standard 12-TET scale used by Western music, by synthesizing naturally harmonic instruments?

Outside of research, I love to make and share social games, video games, computer programs, and music.