A Generalized Linear Model for Estimating Spectrotemporal Receptive Fields from Responses to Natural Sounds
Figure 9
Predictive power of the different models tested.
For both predictions of song and ml noise responses, all the models that use a sparse prior or L1 regularizer (i.e., Exp GLM, Exp LN, spline GLM, and Linear L1) have an average prediction correlation that is significantly higher than the average prediction correlation for NRC, which uses a smoothing prior to regularize the STRFs. We found no significant difference in predictive power across the models that employ a sparse prior. Error bars show SEs.