- Location: Wilson Hall • 111 21St Ave S • Nashville, TN 37240
- Room: 115
- Contact: Angel Gaither
- Email: email@example.com
- Phone: 615-322-0080
- Audience: Free and Open to the Public
Hjalmar Turesson, PhD
Department of Psychology
“Can prior statistical learning facilitate supervised learning? Preliminary experimental results"
Throughout our lives, we are presented with continuous, high-dimensional, streams of sensory data. While we still learn, for example, object categories from this, only occasionally are our interactions with the environment structured so as to allow for learning from a teacher signal. High-dimensional data, but infrequent teacher signals make credit assignment difficult. However, this problem would be alleviated if we learn from all data, both with and without a teacher signal, that is, if learning is semi-supervised. Thus, we address if prior unsupervised learning facilitates later credit assignment while measuring neural correlates of statistical and semi-supervised learning. We will present preliminary results from macaques presented with a visuomotor task of cue-target pairs with different transition probabilities. Initially, as an unsupervised learning task with statistical structure incidental to reward, and later, as a supervised learning with a teacher signal introduced.