Advanced Statistical Modeling for Behavioral and Neural Sciences

Catalog description

This class will teach you how to formulate, train, test, and compare your own custom statistical and mechanistic models for behavioral and neural data. The core of the class is the "universal procedure" of modern modeling that has emerged over the past 10 years, involving: (1) formulating your hypothesis space as a parameterized model, (2) optimizing model parameters to fit data with gradient methods, and (3) fairly evaluating the fitted model using cross-validation. The first part of the class will build understanding by recreating within this framework standard models you may already have encountered, such as regularized linear regression, GLMs, SVMs and logistic regression, linear mixed models, PCA and factor analysis, structural equation modeling, and simple neural networks. The second part of the class will focus on helping you workshop custom models for your own research problems. Prereqs: a working knowledge of Python programming, and Psych 251/253 (or similar courses). A few math tools will be used (derivatives and gradients, and some linear algebra), but we will help you get up to speed on these as part of the class.