Supplementary Figure 3: Empirical quantification of multiplicative effects.

We compare an alternative nonlinearity with (soft rectification/threshold linear function). Each column represents a neuron corresponding to Fig. 5. (a) Exponential function (black) and empirical predicted nonlinearity on test-data (red). Empirical nonlinearities are estimated by first partitioning the net linear output for each time bin into eight equal-quantile buckets, then estimating the expected spike count conditioned on each bucket. The expected means are plotted at the mean net linear drive within each bucket. (b) Threshold linear function (black) and empirical predicted nonlinearity on test-data (red). Note the systematic deviation towards an exponential nonlinearity. (c) To check for multiplicative interactions, we divided the net linear input into two components, one from saccade-locked kernels and the rest. Then, we compared model prediction to that expected from multiplicative interaction between those two components. Each colored trace represents conditioning on different amounts of contribution from the saccade-locked kernels. Solid line is the true multiplicative effect predicted from the model (with exponential nonlinearity), while dotted lines are empirical rate prediction from the data.