I am Joe Austerweil, an Assistant Professor at Brown University in the Department of Cognitive, Linguistic, and Psychological Sciences.
As a computational cognitive psychologist, my research program
explores questions at the intersection of perception and higher-level
cognition. I use recent advances in statistics and computer science to
formulate ideal learner models to see how they solve these problems
and then test the model predictions using traditional behavioral
experimentation. Ideal learner models help us understand the knowledge people use to solve problems because such knowledge must be made explicit for the ideal learner model to successfully produce human behavior. This method yields novel machine learning methods and
leads to the discovery of new psychological principles.
Brown University (2007), Sc. B. in Applied Mathematics-Computer
Science (with honors)
University of California, Berkeley (2011), M.A. in Statistics
University of California, Berkeley (2012), Ph.D. in Psychology
I will be giving a talk at the 47th Annual Meeting of the Society for Mathematical Psychology, which meets from July 18th to July 21st, 2014. The title is Dissociating Top-down and Bottom-up Theories of Autism Spectrum Disorders Using Bayesian Models. It is joint work with Kirstie Stanworth and Anna Franklin. Show/Hide Abstract
We define novel Bayesian models of two cognitive theories of Autism Spectrum Disorders (ASDs) to capture how individuals with ASDs differ from typical individuals when they learn causal relations. People learn causal relations between cues and effects in physical and social domains by integrating how often they have observed the cue and effect together with their domain-specific knowledge about the mechanisms whereby cues generate effects. Prior knowledge allows people to need only a few observations to solve this complex task in both physical (e.g., learn that flipping a switch turns on a particular light) and social (e.g., learn that grabbing a toy from a friend will make the friend cry) domains. The amount of weight given to current evidence when it is integrated with prior beliefs should increase with the strength of evidence, in a manner prescribed by Bayes' rule.
There are many documented social and non-social consequences of ASDs, including effects on learning and the way in which those with ASDs perceive and experience the world around them. Two prominent theories seeking to explain these consequences are a top-down hypothesis, the weak coherence hypothesis, where prior knowledge is underweighted, and a bottom-up hypothesis, the enhanced perceptual functioning hypothesis, where current observations are overweighted. We propose Bayesian models of each theory for the problem of learning a causal relation. The Bayesian models produce divergent predictions, which we tested in a novel adaptation of the "blicket" detector causal learning paradigm. In our modified version, an object's color is a cue to the prior probability that it is a cause (e.g., 80% of red objects are blickets). There were two phases: (1) a familiarization phase, where prior knowledge about the causal relation is established through experience (i.e., how many objects of the current color are blickets), and (2) a test phase, where on each trial, an effect is active, the number of potential causes is manipulated, and subjects rate how probable one of the potential causes is to be a cause (e.g., 32 objects are on an active detector. How probable is it that one particular object is a blicket?). The results support the weak coherence hypothesis: Subjects with ASDs used prior knowledge significantly less than typical participants when learning a causal relation with a small base rate, but otherwise learned causal relations in a manner very similar to typical participants.
I will be giving a talk at the 36th Annual Meeting of the Cognitive Science Society, which meets from July 23rd to July 26th, 2014. The title is Testing the psychological validity of cluster construction biases. Show/Hide Abstract
To generalize from one experience to the next in a world where the underlying structures are ever-changing, people construct clusters that group their observations and enable information to be pooled within a cluster in an efficient and effective manner. Despite substantial computational work describing potential domain-general processes for how people construct these clusters, there has been little empirical progress comparing different proposals to each other and to human performance. In this article, I empirically test some popular computational proposals against each other and against human behavior using the Markov chain Monte Carlo with People methodology. The results support two popular Bayesian nonparametric processes, the Chinese Restaurant Process and the related Dirichlet Process Mixture Model. [Paper PDF]
I will be teaching Thinking (CLPS 1200) in Fall 2014. Syllabus to come soon.
I am interested in taking PhD students for starting graduate school Fall 2015. Please send me an email if you are interested.
Joseph Austerweil and Thomas Griffiths. (2013). A nonparametric Bayesian framework for constructing flexible feature representations. Psychological Review, 120 (4), 817-851. [DOI]
Last Updated May 18, 2014