Introduction#
With modern electrophysiological and optical recording techniques, neuroscientists routinely measure the activity of thousands of cells simultaneously. Coupled with high-resolution behavioral measurements, these datasets offer unprecedented opportunities to learn how neural circuits function. This book covers machine learning methods for analysing such datasets, including:
spike sorting and calcium deconvolution techniques for extracting relevant signals from raw data
markerless pose tracking methods for estimating animal pose in behavioral videos
generalized linear models for studying neural encoding properties
Bayesian decoders for inferring behavior from neural data
point processes for modeling neural spike trains
state space models for analysis of high-dimensional neural and behavioral time-series
sequential variational autoencoders, like LFADS, for modeling neural dynamics.
The book includes both chapters and coding labs. Each chapter develops the theory behind modern models and algorithms for neural data. The chapters are organized into three units:
Signal Extraction: methods for extracting spike trains or pose estimates from raw data
Encoding and Decoding: techniques for relating neural activity to sensory or behavioral covariates
Unsupervised Learning: models for uncovering low-dimensional structure in high-dimensional neural and behavioral data.
In the coding labs, you will hone your programming skills and develop practical understanding by implemeting these methods from scratch (with starter code!) and applying them to real datasets.
This book accompanies a course I teach at Stanford University called STATS 320: Machine Learning Methods for Neural Data Analysis. Unfortunately, video recordings of the lectures are not available online. Likewise, I am not currently releasing solutions to the labs, though I may change that in the future!
The book is a work in progress, but I hope you can already begin to enjoy it. Please reach out with any feedback!