Todd Gureckis, PHD
Department of Psychology, Center of Data Science
The Computational basis of self-guided learning
When I first encountered neural networks in grad school, the idea I could write a simple program that could “program” itself simply by feeding it data was exhilarating. Even more exciting is a program that not only learns from provided data but can actively go out and gather the data it needs to learn. Such a system we might even describe as being “curious” about the world. In my talk, I will review some of the last decade of behavioral research in my lab, exploring the computational algorithms people (including young children) use to actively learn about the world using both actions and language and the consequence that this has on our emotions, memories, and knowledge. A central thesis underlying my work is that we can best fulfill the vision of a fully autodidactic machine learner by carefully studying how humans succeed and fail at guiding their own learning. In the latter part of the talk, I will describe an initial attempt at developing a computational model capable of self-generating playful goals which foster exploration and learning.
Host Information: Yifan Li ([email protected])
The Columbia Neuroscience Seminar series is a collaborative effort of Columbia's Zuckerman Institute, the Department of Neuroscience, the Doctoral Program in Neurobiology and Behavior and the Columbia Translational Neuroscience Initiative, and with support from the Kavli Institute for Brain Science.