Using Matrix and Tensor Factorizations for the Single-Trial Analysis of Population Spike Trains
(A) Examples of RGC spike raster plots in response to shifted natural images. The top row shows the receptive field and raster plots of representative neuron 1. The bottom row shows the same for neuron 2. The two columns to the left are for image 1 and the two columns to the right are for image 2. The scale bar below the image represents the width of the projected image on the retina. The red ellipses indicate the receptive field of the neuron in the presented image. The numbers in the ellipse denote the directions from the center of the ellipse in which the image was shifted. Raster plots show the spike times of the neurons in response to the natural image stimuli in the 9 indicated positions. For better illustration of timing relations we show only the time range from 100 ms to 300 ms after stimulus onset, but analyzed the range from 0 ms to 300 ms. Each row of raster plots shows responses from the same trial. (B-D) Comparison of decoding performance on unshuffled responses with space-by-time NMF (space-and-time), after shuffling bins across time (space-only), and after keeping only the first spike of each neuron in each trial (latency-code). (B) Performance of decoding image id and image position. (C) Performance of decoding image id for each position, averaged over positions. (D) Performance of decoding image position for each image id, averaged over images. ***p<0.001; two-tailed t-test. Error bars indicate s.e.m.