Dr. Chris Daube
16:00 27 July 2022
University of Glasgow | School of Psychology and Neuroscience | Website
Quantitatively comparing predictive models with information theoretic measures
In neuroscience supervised learning methods are increasingly applied to gain insight into neural representations. Frequently, encoding models predicting neural responses are fit to a number of different stimulus features sets. To gain the maximum benefit from this modelling, we need to go beyond simply ranking models by predictive performance. We present information theoretic methods that directly quantify the relationships between predictions of different models, quantifying common vs unique predictive information content. We show how these methods can aid interpretation of both encoding and decoding models in neuroscience, as well as predictive modelling in science more broadly.
