Welcome to the Bonner Lab at the Cognitive Science Department of Johns Hopkins University, led by Assistant Professor Mick Bonner. Our research focuses on the intersection of neuroscience and artificial intelligence. We want to understand the fundamental principles of computation in the brain, and we study them by asking what brains and deep network models have in common — how each learns to represent the visual world. Our approach combines large-scale neuroimaging with computational modeling, and we analyze biological and artificial systems with the same set of tools so that they can be compared directly.
Our recent work shows that networks with very different architectures and training objectives converge on a common set of representational features, and that these universal features are the ones most reliably found in visual cortex. We are now asking where such representations come from: how much alignment with the brain arises from architecture and image statistics alone, from extremely coarse feedback, or from local learning rules that build a visual hierarchy without backpropagation. Underlying all of this is our finding that neural representations are intrinsically high-dimensional, which sets the bar for what a model of the brain has to explain.
The lab regularly attends the Cognitive Computational Neuroscience (CCN) and Vision Sciences Society (VSS) meetings, and occasionally NeurIPS. If you are interested in this research, feel free to reach out to anyone in the lab.
