Fig 3. Performance of stimulus-restricted and run-restricted LMs in predicting single trials.

(A) Top, stimulus-dependent average response profile for an anecdotal example neuron. The average response is a grand mean, i.e. the average response of a neuron to a given stimulus is the neuron’s response to the stimulus averaged across all time bins of all presentations of the given stimulus. In effect, the average response is the average fluorescence change the neuron exhibits in response to a given stimulus. Bottom, an example of a block averaged trace for the same neuron (neuron 2, dataset 3). The block averaged trace of a neuron for a given stimulus is the trace obtained when averaging traces across all blocks (i.e. trials) and all presentations of a given stimulus within each block. In effect, the block averaged trace represents the average trace of fluorescence changes in response to a given stimulus. For the LMs using the block averaged trace model, we restricted our analysis only to the stimulus frames and excluded the grey frames which are obscured here using grey bars. (B) Left, cumulative distribution functions of the percent change of test set MSE across stimulus frames for the stimulus-restricted LM using the average response stimulus model and the stimulus-restricted LM using the block averaged trace stimulus model. Right, zoom in on 95^th percentile. (C) Cumulative distribution functions of the percent change of test set MSE across all frames for the run-restricted LMs using the rotary encoder running model, nonnegative model coefficients, and average population response running model.