Figure 7.

The CRNN models with different recurrent layers

(A) Structures of different kinds of recurrent units.

(B) Spatial RFs and temporal filters of neurons learned by the CRNN with vanilla RNN, GRU, and LSTM.

(C) Performance of the CRNN models based on different recurrent units to predict the neural response to two natural video stimuli (movie 1 and movie 2). Population CC, the correlation coefficient between population RGCs and the average output of the models; average CC, the correlation coefficient between the RGCs and the models' outputs.

(D) RF distance of each cell between the data and the model outputs.