Dan Scott

Doctoral Student
B.A. Physics & B.A. Mathematics
University of California | Berkeley, California 

Research Interests 
My research program at Brown revolves around reinforcement learning, computational psychiatry, and network computation. If I have one over-arching goal at the moment, it is to understand how aberrant network function produces changes in psychiatrically relevant processes. Reinforcement learning is a relatively well studied system in which variations in function at multiple levels of analysis are known to produce many interesting effects, and as such constitutes an excellent model system for these types of investigations. That aside, it's also fascinating in its own right, can be approached normatively, computationally, algorithmically, etc, and has far reaching connections in mathematics, statistics, and physics.

Prior Work 
At UC Berkeley I worked with a graduate student in Tom Griffiths’ Computational Cognitive Science Lab to extend Tom’s “causal support” model of causal induction, which formulates the task of inferring causal relationships as a Bayesian inference problem over the structure and edge weights of directed acyclic graphs. The causal support model was originally defined with respect to Boolean data in discrete time, so I worked with Mike Pacer (a grad student in the lab) to extend its scope to include continuous data in continuous time.

Graduating from Berkeley, I wanted more exposure outside of math and physics. I thought I'd like to go to graduate school but my other primary 'applied' interest, climate change, was something I still knew relatively little about from a practice-of-science standpoint. To pursue this interest, I moved to Cambridge and took a research assistantship at Harvard with Paul Moorcroft of the Moorcroft Lab. While there, I worked on a computational model of the terrestrial biosphere called the Ecosystem Demography Model, which predicts pools and fluxes of carbon, water, and energy within and across the boundaries of land-based ecosystems. My primary goal in this work was updating the model's carbon accounting subsystem and doing so has resulted in a significant improvement in the model's fit to available data. In the process of accomplishing this goal I was very lucky to have worked with many collaborators at universities and national labs, as well as Paul himself, and the model - with hundreds of thousands of lines of Fortran and multiple levels of parallelization requiring expensive-black-box optimization - was a great entree into high performance distributed computing, software engineering, multi-institution code-bases, and numerous numerical and computational methodologies. The one down-side of this work was realizing that I'm just not as passionate about the fundamental questions of earth system science as I am about neuroscience, but this was important for me to learn.

Cognitive neuroscience never got less interesting however, and given my background, Michael's lab here at Brown seemed like the perfect place to hash out my multi-level modelling interests, particularly as they relate to improving peoples' lives and our collective understanding of basic human behavior.

Contact Information
Email: daniel [underscore] scott [at] brown [dot] edu
Hometown: Orinda, CA