Research Statement Angela Yu

My long-term goal is to develop a theoretically and neurobiologically grounded understanding of different facets of cognition that give rise to intelligent behavior.

My approach

Using a multi-disciplinary approach combining theoretical and experimental methods, my research aims to identify the computational principles and neural basis of key cognitive functions, such as vision, attention, learning, cognitive control, decision making, active learning/sensing, economic choice, and social cognition.

Mathematical models

This means having appropriate mathematical models of the behavioral tasks faced by the brain, the forms of data available to the brain, and the actions that the brain can take to achieve behavioral goals, as well as the nature of information representation and processing required to support the necessary computations.

Research questions

In the parlance of Marr’s three levels of analysis (Marr, 1982), my approach is integrative and comprises all three of “computational”, “algorithmic”, and “implementational” levels of analyses. I aim to answer questions such as: How is information represented, computed, and utilized in the brain in order to support intelligent behavior? What neural and computational processes go awry in various disorders of the mind and the brain (e.g. psychiatry and neurology)?

Bayesian Modeling

Conceptually and methodologically, my work is related to a growing body of Bayesian statistical modeling work showing that various aspects of perception and learning, both its features and apparent bugs (such as visual illusions) can be understood as statistically normative (optimal) inferences about the world based on noisy data. Although the brain is unlikely to be exact-Bayesian (optimal) in general, Bayesian models provide a principled formalism from which to build computational and algorithmic understanding of brain processes. However, Bayesian statistical inference alone does not provide adequate formal tools to explain many aspects of cognition; in particular, it is chiefly concerned with the representation, integration, and propagation of inexact information, but it is silent on what course of actions or decisions ought to be taken based on that information, in order to achieve a desired behavioral outcome. As such, Bayesian analyses have had the most success in modeling aspects of cognition that only require observational analysis of incoming data, such as in perception and learning, and not those that require active manipulation of the state of the world or the nature of the incoming data. To overcome this limitation,

Decision, Control and Game Theory

I have been combining Bayesian modeling with decision theory, which deals with how to optimize terminal actions, control theory, which deals with how to optimize actions that affect future data collection (as in active sensing) and/or the state of the world, and game theory, which deals with strategic social interactions.

Methodologically, my lab typically uses behavioral data to constrain and develop the initial computational model, and then seek collaborations with neurophysiologists and psychiatrists to identify the underlying neural substrate that support the computations implied by the models in the healthy brain, as well as understanding the etiology and manifestation of impairments to component processes in psychiatric and neurological diseases.

Strands of Research

Human Face Processing
Human Face Processing
Active Cognition
Active Cognition
Rational Irrationality
Rational Irrationality
Inhibitory Control
Inhibitory Control
Neuromodulation, Learning and Attention
Neuromodulation, Learning and Attention
AI and Machine Learning
AI and Machine Learning

Future Directions

Future Directions

To achieve a theoretical understanding at many different levels (from neurons to cognition to behavior), research projects in my lab typically tend to span many years, as the theoretical insights obtained along the way inform the next step of neuroscientific experimental inquiry, and as I gradually gain more experimental collaborators.

Learn more