UIUC Quant Brownbag

Psychological theories are commonly implemented in latent-variable models that predict and explain behavior in terms of unobserved traits and processes. These models can be difficult to develop and test, especially when they lack convenient likelihoods and have to rely on simulation instead. More generally, computational models can also be difficult to use for those who lack extensive training in computational and statistical methods. In this talk, I examine several avenues where artificial intelligence (AI) and machine learning (ML) can benefit modeling in psychology by providing new methods for developing, testing, and disseminating models. First, machine learning can be used to efficiently estimate the values of latent parameters and their (Bayesian) posterior distributions. Second, machine learning classifiers can be used to assign probabilities to competing models, improving on existing approaches to model comparison. Third, exploratory data analysis techniques like variational autoencoders can be used to identify latent dimensions when there are no existing models of a particular measure. Finally, these tools allow models to be embedded within simple point-and-click user interfaces, making them accessible to a broader audience.