Research

We’re looking for an expert computational modeler for a postdoctoral fellowship position to start Summer/Fall 2019. Click here for more information.

The lab’s research focuses on understanding information processing in human learning, decision making, and episodic memory, and how it is disrupted in psychiatric disorders (anxiety, obsessive compulsive disorder, and depression) – an area recently dubbed computational psychiatry. In carrying out this work, we leverage tools from cognitive neuroscience, experimental and mathematical psychology, computer science, and statistics. We work at the intersection of theory and experiment, building computational models of information processing and testing them using both behavioral data and neuroimaging (fMRI and scalp EEG).

Current Themes

The temporal dynamics of multi-step decision making

Imagine deciding whether you should go to graduate school, whether you should move somewhere for work, or what route to take to your new job. Everyday decisions necessarily involve thinking through multiple steps of action. However, we know surprisingly little about how the brain is able to accomplish this important task. In particular, we are interested in the temporal dynamics by which multi-step decisions unfold: not only what choices people make, but what the process of deliberation looks like. We previously proposed and tested several alternative models of planning (Solway & Botvinick, 2012, 2015). However, many questions remain, including:

  • What assumptions of the current ‘best’ models are reasonable, and which are not?
  • What are the neural correlates of the dynamics predicted by these models?
  • What new experimental paradigms and manipulations challenge current models and provide new modeling constraints? What changes need to be made to existing models to accommodate these challenges?
The interaction between habitual (“model-free”) and goal-directed (“model-based”) control

Classical notions of habitual and goal-directed control have recently been formalized within the framework of reinforcement learning. Habitual, or model-free, control refers to decision making based on stimulus -> response or stimulus -> value mappings built up through repeated experience. Goal-directed, or model-based, control on the other hand is more flexible and prospective, building up decisions in real-time by combining knowledge about the environment’s local transition and reward structure (the “model”). The last decade has seen an explosion in work trying to understand how each decision system works, and how the balance between these systems differs as a function of a multitude of individual differences. However, we still know very little about how the two systems actually interact to produce a single final decision. We are studying the computational and neural properties of this process.

How is decision making disrupted in obsessive-compulsive disorder?

The long-term prognosis of obsessive compulsive disorder remains poor for a large percentage of patients even after treatment, and symptoms can often be debilitating. Making substantial progress in treatment effectiveness requires a better mechanistic understanding of this disorder. We are using computational modeling, in tandem with neuroimaging, to better understand information processing differences. In particular, questions we are interested in include:

  • How do decision making and memory differ exactly, and how do these differences interact with previously reported impairments in executive functioning?
  • Are there domain general differences in information processing?
  • Are there different symptom clusters that map on to different changes in this highly heterogeneous disorder?