Speaker: John Morrison (Columbia University)

Title: Using transfer learning to identify a neural system’s algorithm

Abstract: Algorithms transform inputs into outputs through a series of operations on intermediate representations. Cognitive psychologists use algorithms to explain cognition. For example, they use tree-search algorithms to explain planning, reinforcement learning algorithms to explain exploration, and Bayesian algorithms to explain categorization. There is disagreement among systems neuroscientists and machine learning researchers about whether and how to use these same algorithms to explain neural systems. A standard method is to look for parts in the neural system corresponding to the parts of the algorithm, thereby treating the algorithm as a causal model of the neural system. However, this method has struggled to find the algorithms used in cognitive psychology, leading some to view these algorithms as merely useful fictions – useful for predicting a neural system’s outputs but fictional as causal models of its operations. Others view these algorithms as merely normative, indicating the ideal input-output mapping without attributing any particular operations to the neural system. As an alternative method, we suggest identifying a neural system’s cognitive psychology algorithm by assessing how quickly it learns alternative input-output mappings, that is, its transfer learning profile. The basic idea is that different algorithms make different input-output mappings easier to learn allowing us to recover a neural system’s original algorithm from its transfer learning profile. We use artificial neural networks to demonstrate that this method productively applies to a range of neural systems on a range of tasks. We conclude that transfer learning is a promising approach for integrating cognitive psychology algorithms with neural systems and thus for integrating cognitive psychology with systems neuroscience and machine learning.

Biography: I am a professor of philosophy at Barnard College, Columbia University. I am an affiliate of Barnard’s Neuroscience and Behavior Department, Columbia’s Mind Brain Behavior Institute, and Columbia’s Center for Theoretical Neuroscience. I am also a mentor in Columbia’s Neurobiology and Behavior Graduate Program. My research has been supported by the National Endowment for the Humanities, National Science Foundation, Mellon Foundation, and Data Sciences Institute. I am an editor of the Journal of Philosophy. I led the effort to create the cognitive science program at Barnard and Columbia, and served as Barnard’s founding director.