Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Dataset Distillation with Convexified Implicit Gradients

About

We propose a new dataset distillation algorithm using reparameterization and convexification of implicit gradients (RCIG), that substantially improves the state-of-the-art. To this end, we first formulate dataset distillation as a bi-level optimization problem. Then, we show how implicit gradients can be effectively used to compute meta-gradient updates. We further equip the algorithm with a convexified approximation that corresponds to learning on top of a frozen finite-width neural tangent kernel. Finally, we improve bias in implicit gradients by parameterizing the neural network to enable analytical computation of final-layer parameters given the body parameters. RCIG establishes the new state-of-the-art on a diverse series of dataset distillation tasks. Notably, with one image per class, on resized ImageNet, RCIG sees on average a 108\% improvement over the previous state-of-the-art distillation algorithm. Similarly, we observed a 66\% gain over SOTA on Tiny-ImageNet and 37\% on CIFAR-100.

Noel Loo, Ramin Hasani, Mathias Lechner, Daniela Rus• 2023

Related benchmarks

TaskDatasetResultRank
Image ClassificationImageNet I-Squawk (test)
Accuracy49.9
71
Image ClassificationImageNet-A (val)
Accuracy25.4
64
Image ClassificationImageNet-Woof (test)
Accuracy32.9
46
Image ClassificationImageNet I-Fruit (test)
Accuracy35.3
23
Image ClassificationImageNet I-Yellow (test)
Accuracy53.8
22
Image ClassificationImageNet B 2012 (val)
Estimated Accuracy21.3
17
Image ClassificationImageNet Meow (test)
Accuracy37.1
16
Image ClassificationImageNet-Nette (test)
Accuracy40
16
Image ClassificationImageNet Subset E (val)
Accuracy17.1
9
Image ClassificationImageNet Subset C (val)
Accuracy21.2
9
Showing 10 of 11 rows

Other info

Follow for update