Modern Machine Learning as a Benchmark for Fitting Neural Responses

Abstract

Neuroscience has long focused on finding encoding models that effectively ask “what predicts neural spiking?” and generalized linear models (GLMs) are a typical approach. It is often unknown how much of explainable neural activity is captured, or missed, when fitting a model. Here we compared the predictive performance of simple models to three leading machine learning methods: feedforward neural networks, gradient boosted trees (using XGBoost), and stacked ensembles that combine the predictions of several methods. We predicted spike counts in macaque motor (M1) and somatosensory (S1) cortices from standard representations of reaching kinematics, and in rat hippocampal cells from open field location and orientation. Of these methods, XGBoost and the ensemble consistently produced more accurate spike rate predictions and were less sensitive to the preprocessing of features. These methods can thus be applied quickly to detect if feature sets relate to neural activity in a manner not captured by simpler methods. Encoding models built with a machine learning approach accurately predict spike rates and can offer meaningful benchmarks for simpler models.

Machine Learning for Neural Decoding

“…The same task was used in the recording from somatosensory cortex ( Benjamin et al, 2018 ). The recording from S1 was 51 min and contained 52 neurons.…”

Capturing Multiple Timescales of Adaptation to Second-Order Statistics With Generalized Linear Models: Gain Scaling and Fractional Differentiation

“…Additionally, realistic input statistics may not follow a Gaussian distribution (Heitman et al, 2016;Maheswaranathan et al, 2018). Further work toward understanding the adaptive computations performed by single neurons should consider the inputs the neuron receives within a broader network and should consider non-linear stimulus processing (McFarland et al, 2013;Benjamin et al, 2018). Neural coding and computations that occur across a wide range of input levels depend heavily on adaptation to the stimulus variance (Wark et al, 2007).…”

Capturing multiple timescales of adaptation to second-order statistics with generalized linear models: gain scaling and fractional differentiation

“…Predictive performance was evaluated using the pseudo- R 2 score (Cameron and Windmeijer, 1997). We selected this measure because it can be applied to Poisson process observations instead of trial-averaged firing rates as is required by the standard R 2 measureof explained variance (Benjamin et al, 2018). Thus, it is especially appropriate for comparing the stochastic GLM to a spike train simulated by the deterministic HH model.…”