Yudi Xie

Yudi Xie

PhD Student

MIT

How does the brain learn? How to apply our knowledge of how the brain learns to help people learn, design intelligent machines, and in a broader sense, benefit the entire society? My research is motivated by these questions.

My research interests are at the intersection of machine learning and neuroscience. There are many kinds of learning in machine learning terms, such as supervised learning, unsupervised learning, and reinforcement learning. Previous research has found links between these kinds of learning and biological processes in the brain. However, people mainly studied different kinds of learning separately. I am working on building integrated multi-systems models that account for how different kinds of learning happen and interact in distributed processes in the brain. Several key components of learning are identified:

  1. fast, few-shot learning and inference in the prefrontal areas,
  2. slow but important procedural learning in the basal ganglia,
  3. unsupervised learning in the cortex,
  4. supervised learning in the cerebellum,
  5. and a motivation system generate goals and reinforcement that facilitate learning.

Eventually, I hope to build a unified model that combine all these components to account for all kinds of learning processes in the brain. I will use the model to generate testable predictions about learning behavior in human and animal. The model will be compared and validated by neural recoding data. Besides the brain, the model will provide important insights about how to build intelligent machine, for example providing insights to hierarchical reinforcement learning and the value alignment problem in AI.

Interests
  • Learning
  • Motivated Behavior
  • Decision Making
  • Computational Neuroscience
Education
  • PhD in Brain and Cognitive Science, 2019, 2021 - Present

    Massachusetts Institute of Technology

  • BS in Physics, 2015 - 2019

    University of Science and Technology of China