A bit of a background
Back in September 2023, after the Bernstein Conference, I came back from Alex Cayco-Gajic’s workshop talk quite inspired. The low tensor rank framework she presented (work together with Arthur Pellegrino and Angus Chadwick Pellegrino, Cayco Gajic, and Chadwick (2023)) had struck me as one of the more conceptually elegant approaches to track how neural population dynamics change over learning. I very enthusiastically shared my excitement in the internal post-Bernstein group meeting, as one does when ideas resonate1.
It turned out that sharing my excitement about someone else’s work2 didn’t land equally well with everyone in the room, but that’s life. The dynamics that followed are perhaps best saved for a more informal conversation.
A few months later, in December 2023, I noticed the call for the ICLR 2024 blogpost track, and thought this would be a great opportunity to spotlight this work. I began drafting a piece, but for a mix of personal and political reasons (and admittedly, some competing deadlines), I set it aside3.
Since I am still excited about this work, stay tuned here, I will soon revive this post.
In the meantime, I recommend going straight to the source, it’s a great read.
