Paper out in journal Science
New paper out in print: 'Efficient coding explains the universal law of generalization in human perception'. In this paper I show that the "universal law of generalization" emerges inevitably from any information processing system (whether biological or artificial) that minimizes the cost of perceptual error subject to constraints on the ability to process or transmit information. Link to publications page.
Rate-distortion theory and human perception
New paper out in print: 'Rate-distortion theory and human perception'. This paper provides a tutorial introduction to a theoretical framework for understanding perception and perceptual memory based on a branch of information theory. Link to publications page.
The Cost of Misremembering
New paper out in print: 'The cost of misremembering: Inferring the loss funciton in visual working memory'. This paper applies information theory and decision theory to understand visual working memory in an entirely new light. Link to publications page.
The Adaptive Nature of Visual Working Memory
New paper out in print: 'The Adaptive Nature of Visual Working Memory'. This is an accessible overview of how statistical learning plays a key role in visual memory, and highlights the crucial role of computational theory in psychology. Link to publications page.
Adaptive Computational Cognition Lab
Our laboratory is interested in answering the following question:
How is the brain able to accomplish complex goals, with limited computational resources, in an uncertain world?
In addressing this question, our research encompasses a broad range of topics, including visual perception, memory, motor control, and reinforcement learning. In each case however, we apply a common theoretical framework. Simply stated, the goal of information processing in the brain is to transform information into action. An optimal cognitive system is one that maximizes the utility of action, subject to constraints on the ability to store and process information.
As an intuitive example, when reaching to pick up a cup of coffee, your visual system must extract features relevant for the task, such as the depth of the cup in space. This information is stored and manipulated in visual working memory, and must be combined with a control policy that maps estimated states of the body and world onto motor commands. Critically, each one of these stages must be carried out in a biological system with finite information processing capacity. Hence, an optimal algorithm for performing this task on a system with unlimited computational resources will be very different than the optimal algorithm for a biological system.
To test this idea, we conduct behavioral experiment on human learning, perception, and memory. Our laboratory is equipped with state of the art technology for mobile eye tracking and full-body motion capture that allows us to measure and record behavior in fine-grained detail across a range of tasks. To understand human performance in these laboratory tasks, we develop computational cognitive models that derive from machine learning, Bayesian statistics, and information theory.
In everyday life, the incredible complexity of the problems solved by the brain is hidden from our awareness. Oftentimes, it is only when we try to replicate the breadth and robustness of human performance in a computational model is the extent of the mind's complexity apparent. Computational models of human cognition serve an important role in all of the research carried out in our laboratory. They serve as an explicit statement of a scientific theory, but they can also generate novel predictions that can be tested experimentally. In many cases, human performance exceeds that of any existing algorithm (for example, in categorizing images or recognizing faces). By studying human behavior and building computational models of cognition, we can also advance the state of the art in engineering and applied sciences.
If you are interested in becoming involved in our research, please feel free to contact us.