Continual Learning

UCSD CSE & Salk Institute Student Researcher

The modern paradigm of machine learning puts the emphasis on the end result, rather than the learning process, and overlooks a critical characteristic of human learning: that it is robust to changing tasks and sequential experience.

I am currently researching into the theoretical aspects of Continual Learning and Transfer Learning with advising from professor Sichun Gao from UCSD Computer Science & Engineering Department and professor Talmo Pereira from Salk Institute. I try to frame the general problem of Continual Learning from the perspective of Reinforcement Learning & Cognitive Neuroscience, hoping to develop algorithms that utilize the same strategies of "how we learn" onto an artificial agent.

Schematic of Transfered Learning (Zhu et al. (2023))

I am interesting in developing a conceptual framework for continual learning that may be used for any families of algorithms and I'm currently studying it through the scope of control problems in reinforcement learning and trying to build internal representations of physics for artificial agents so their learned skills in one task can be modularized and transferable to other control tasks.