Tutorial

As part of the workshop on generative models, there will be a tutorial, during which we will give a brief overview and examples of some of the generative modeling approaches that will be covered in other talks at the workshop. In addition to the live tutorial, these examples can be found on the github:

https://github.com/davindicode/cosyne-2023-generative-models

This tutorial is divided into three sections. First, we present a series of supervised models ranging from linear regression to Gaussian process regression. In these models, it is assumed that we have access to some set of regressors (e.g. behavior) and want to predict an unknown variable (e.g. neural activity) on the basis of some training data where we get to observe both. We then move on to the latent variable setting, where we have to infer both the regressors (now latent states) and the function mapping them onto an observed variable (e.g. neural activity). Finally, we consider latent variable models with discrete state spaces, which are becoming increasingly popular in the neuroscience community to describe e.g. different behavioral or motivational states.

For these examples, we have not coded all methods for scratch but instead use established libraries in the field. Indeed, the purpose is not to provide a detailed implementation of the methods but rather an overview of how they relate to one another. We hope that these notebooks can help people become more familiar with the range of generative models commonly used in neuroscience, and we welcome any additional questions and contributions.

Questions and comments can be addressed to mmcs3@cam.ac.uk, ktj21@cam.ac.uk, or dl543@cam.ac.uk.