PhD student, Massachusetts Institute of Technology
1 paper at NeurIPS 2025
We distill large datasets to just one image per class and use them to train linear probes on top of self-supervised vision models.
