Date of Award

Fall 1-1-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Physics

First Advisor

Murray, John

Abstract

The ability to learn from experience is essential for both biological and artificial agents. To learn efficiently in complex, high-dimensional environments, agents must generalize strategically from limited data by focusing on features relevant for the task. The generalization strategy, known as inductive bias, shapes the dynamics of learning. Previous works have established the theory of neural tangent kernels that describes the learning dynamics of artificial neural networks. However, a comparable framework for understanding and quantifying inductive bias in biological systems performing behavioral tasks remains underdeveloped. Here, we introduce a neural kernel framework to characterize inductive biases of humans and artificial neural networks in category learning tasks, thereby linking neural representations with learning behavior. Our kernel models captured the learning trajectories of human subjects across two experiments, and elucidated the learning strategies of humans and neural networks using feature modes. We developed novel methods for fitting kernels to behavioral data. The fit kernels not only revealed the adaptation of inductive bias across multiple tasks, but also predicted how feature representations would change during learning. Furthermore, we implemented a neural network model with feature-based gain modulation, which recapitulated inductive bias adaptation in humans by adjusting feature representations based on task demands. In summary, we have established a comprehensive framework for understanding learning and generalization in relation to neural representations, providing testable predictions for future studies on the neural mechanisms of inductive bias adaptation.

Download

Share

COinS