Figure 2.

Low-rank MNE, full-rank MNE, and STC models were optimized on a dataset of 50 avian auditory forebrain neurons. Logical OR and logical AND FB models were fit using linear combinations of the subspace components of each method. Logical AND FB models from two example neurons are shown (A,B). The quality of each model is measured using the difference between the mean negative log-likelihood of the model and the linear MNE model evaluated on the test sets and plots summarize predictive ability across the population of neurons (C). A bar plot quantifies the number of neurons in the population best fit by each model (D). Note low-rank and linear MNE models outperform all STC and full-rank MNE models across the population.