Seems cool, but the image classification model benchmark choice is kinda weak given all the fun tools we have now. I wonder how Tversky probes do on top of DINOv3 for building a classifier for some task.
no sense spending large amounts of compute on algorithms for new math unless you can prove it can crawl.
It's also a more natural question to ask, since building projections on top of frozen foundation model embeddings is both common in an absolute sense, and much more common, relatively, than building projections off of tiny frozen networks like a ResNet-50.