Fig. 3

Firing-rate adaptation, neural response variability, and population coding. The average normalised population firing rate during luminance-increment (a), luminance-decrement (c), contrast-increment (b), and contrast-decrement (d) switches. Each PSTH was normalised relative to a shuffled PSTH, generated by shuffling the spike times. The shuffling process was repeated 50 times while the spike times across the entire 60-min recorded data were randomised in each run. For visualisation, PSTHs have been convolved with a Gaussian window (50 ms). e–h As in (a–d), but for Fano factor. i Decoding accuracies early and late following each of the eight luminance–contrast switches. Each data point shows the average and standard deviation of decoding accuracy for one switching condition (averaged over 15 runs). Solid points indicate upward switches; empty points indicate downward switches. The black line connects data points associated with low vs high luminance, when the contrast is high. The inset demonstrates the colour code for each luminance and contrast switch. Decoding algorithm: linear discriminant analysis (LDA); the number of neurons = 50 randomly selected out of 390 neurons; width of spike-counting window  =  15 ms; and the number of random runs = 15. Error bars are standard deviations. Support vector machine (SVM) classifier provided very similar results. PSTH, peri-stimulus time histogram