The main focus of our research is understanding how the brain makes sense of complex scenes in the environment, resulting in perception and cognition. We are building computational neural models based on the appealing hypothesis that images, movies, and sounds have predictable and quantifiable statistical regularities to which the brain is sensitive. We expect the scene statistics framework to provide insights both about what the system is computing, why it's computing what it is, and to constrain mechanistic models about how the system implements the computations. We have developed approaches for understanding nonlinearities in cortical processing, including the influence of spatial context (what surrounds a given object or feature) and temporal context (what has been observed in the past). Moreover, we have been pushing the notion of using scene statistics to make testable predictions about how cortical circuits in the brain processes natural scenes. In contrast, many studies in the past, including our own, have derived models based on scene statistics principles, but tested the models only with simple stimuli such as gratings or bars.

Using vision as a paradigmatic example, we are currently particularly interested in understanding how the brain processes visual information hierarchically, to build up more complex representations. We are combining recent advances in deep learning, with approaches we have developed for modeling nonlinearities in Primary Visual Cortex, to make headway in understanding Secondary and higher areas of visual cortex. To this end, we are developing both hierarchical unsupervised learning models based on scene statistics, and incorporating nonlinearities into deep convolutional neural networks trained on object recognition. We expect that developing hierarchical models with more biologically motivated nonlinearities will also lead to better artificial systems, that have perceptual and cognitive abilities more like humans, and can generalize better across stimuli and tasks. We are also working on collaborative projects with the medical school, for instance using machine learning to help find treatment for brain injury.