New publication in Nature Communications on modeling brain representations of object context using the co-occurrence statistics of vision and language. Work with The Epstein Lab. Here's what we found... nature.com/articles/s4146…

New preprint shows how tuning for mid-level features gives rise to selectivity across multiple levels of complexity: from categories, such as scenes and landmarks, to surprisingly simplistic stimuli, such as oriented edges and boxy shapes. w/ Donald Li biorxiv.org/content/10.110…

Excited to announce that my first paper as senior author: “Finding Biological Plausibility for Adversarially Robust Features via Metameric Tasks” has been accepted to ICLR (Spotlight scores 8,8,8,8) with lead of former undergrad Anne Harrington CBMM: openreview.net/forum?id=yeP_z…

Memory is not like a camera; it often fills in extra visual details. In a new paper Association for Psychological Science, we (Shreya Wadhwa & Mick Bonner) use "tilt-shift" to explore what causes distortions in scene memory—and find that spatial scale plays a crucial role! psyarxiv.com/hy3qs

Why are deep neural nets (DNNs) so good at modeling the brain? Our new paper with Mick Bonner reveals a striking geometric explanation for DNN models of visual cortex: performance scales with the latent dimensionality of a DNN's natural image manifold. biorxiv.org/cgi/content/sh…

I've been meaning to write this post for a long time: what does it mean for a neural network to be like the brain? I get into the nitty gritty of scores that compare a neural net vs. the brain. xcorr.net/2023/04/20/how…

New paper out in PLOS Comp Biol. The best deep neural network models of visual cortex do not reduce representations to low-dimensional manifolds—instead, they benefit from high-dimensionality. Led by a fantastic student, Eric Elmoznino. journals.plos.org/ploscompbiol/a…

Excited for the lab's presentations VSS Meeting #VSS2024!

Standing room only at #phiVis! So encouraging to see this much interest in the intersection between philosophy and vision science. #VSS2024 VSS Meeting

Why do varied DNN designs yield equally good models of human vision? Our preprint with Mick Bonner shows that diverse DNNs represent images with a shared set of latent dimensions, and these shared dimensions turn out to also be the most brain-aligned. arxiv.org/abs/2408.12804

Excited for #VSS2025! Check out this cool work from the lab.