Self-contained code to demonstrate how to use mTDR. An analysis framework for of trial-structured neural population data using a combination of regression and dimensionality reduction.
This MATLAB code is a reference implementation for the analyses found in Aoi, Mante, and Pillow 2020.
-
From command line:
git clone [email protected]:pillowlab/mTDRdemo.git -
In browser: click to Download ZIP and then unzip archive
Open mTDRdemo.m to see how we estimate the parameters and the number of dimensions by AIC. demoLearning.m is a more step-by-step view of the functions used for parameter learning when the number of dimensions have been specified.
Suppose we record spike responses from a single neuron during a complex behavioral experiment, and would like to know what aspects of the stimulus or behavior are encoded in the neural response. This code package allows us to discover such dependencies using Poisson GLM regression.
Consider a simple example in which a neuron encodes two experimental variables: the time at which a visual target appears, and the motion strength of a moving-dots stimulus. The regressors are the time at which the targets appear, and the time, duration, and strength ("coherence") of the moving dots on each trial.
- MC Aoi, V Mante, & JW Pillow (2020). Prefrontal cortex exhibits multidimensional dynamic encoding during decision-making Nature Neuroscience 2020.