John C. Raisbeck

Ph.D. Candidate
University of Massachusetts Amherst

I am a Ph.D. student in the Manning College of Information and Computing Sciences at the University of Massachusetts Amherst. Before beginning my Ph.D., I worked at the Center for Systems Biology at Massachusetts General Hospital under Professor Hakho Lee. I received my B.S. in Mathematics from the University of Massachusetts Lowell.

My research interest is in the phenomenon of machine learning: how machines learn, how different machines learn differently, and the sets of possible solutions. I view learning through the lens of Reinforcement Learning, the "most general subfield" of machine learning. For more, please see my research statement.

I am always looking for collaborators and interested in working with motivated students.

Papers

Partial derivatives and update direction, built one parameter at a time

Learning & Remembering

Separating learning into a learning rule and a memory rule, and showing that the alignment between the two governs how much an agent can improve.

AltNet

Combatting plasticity loss by switching between an active and a passive network, allowing full resets without a performance drop.

Agent Spaces

A descriptive topology on the set of agents in a decision process, providing a new footing for exploration in infinite problems.

ES Converges to Finite Differences

A proof that Evolution Strategies and finite differences gradient approximators converge as dimensionality grows.

A Scalable Finite Difference Method for Deep RL

Combining the ES/FD result with agent-space Novelty Search into a fully parallelizable RL method.

Writing

Not So Gradient Free

An argument that the term "gradient-free" is deceptive and should be retired in favor of more descriptive language.

Read

Getting Into Research

For students taking their first steps into research: what to expect, how to find an advisor, and how to approach one.

Read