We often train neural networks on classification tasks. After training, what do the network outputs really mean? If they are probabilities of certain categories, then what kind of probability are they? I wrote a blog that tries to explain this: yudixie.github.io/blog/2024/inte…

Very happy to share a new paper that will appear in UniReps proceedings (unireps.org/2024). I show how two popular neural network comparison methods - CKA and RSA - can sometimes be equivalent, i.e. produce the exact same value. links below 👇

Chapter 1 of my book is now available for (free!) preview Princeton University Press. When writing it, I did not anticipate how important ideas about how science will meet societal needs would become in 2025. I have many thoughts - both critical, and optimistic. press.princeton.edu/books/hardcove…

Excited to share the article from MIT News covering our recent work. Huge thanks to everyone who made this possible. news.mit.edu/2025/visual-pa…

If you've wondered how to interpret the outputs of a network trained for classification, come to my other poster at #ICLR2025 tomorrow, where I'll present a post I wrote for the ICLR blog post track. Time and location: Sat Apr 26, 10am - 12:30pm, Poster session 5, poster #335.

Come and visit my poster tomorrow at #ICLR2025 ! I’ll be presenting our research on representation learning and vision. We found that vision models trained on spatial tasks can also be good models of the ventral visual stream. Sat Apr 26, 3-5:30 pm, Poster session 6, poster #65