What are the mechanisms of artificial perception? My work aims at the establishment of a “machine neuroscience” — an integrated theory and set of methodologies for understanding (1) how deep neural networks represent the world, (2) how different model design decisions shape these representations, (3) how model internals explain output behavior (e.g. failure modes), and (4) how to achieve representations & internal dynamics that result in desirable behavior (e.g. adversarial robustness or out-of-distribution generalization). I believe that such an understanding will serve as a bedrock for the development of trustworthy, robust, & general artificial intelligence. 

​

Here are some of the questions I’ve been thinking about: