Method · Autoregressive and flow-based synthesis
ProtoVAR
ProtoVAR: Efficient Dataset Distillation via Prototype-Guided Visual Autoregressive Modeling
Mingyu Wang, Wei Jiang
ICML 2026 · first public 2026-01
In one paragraph
Replaces diffusion sampling with a visual autoregressive (VAR) model guided by multi-scale class prototypes, adding a fast filtering stage that keeps only the most informative generated samples; reported to run up to 70x faster than diffusion-based generative distillation while remaining competitive on ImageNet-scale benchmarks.
Where it sits
- Autoregressive and flow-based synthesis (Generative priors)
- Setting: Image classification
BibTeX (generated; prefer the venue's official entry)
@article{wang2026protovar,
title = {ProtoVAR: Efficient Dataset Distillation via Prototype-Guided Visual Autoregressive Modeling},
author = {Mingyu Wang and Wei Jiang},
journal = {ICML 2026},
year = {2026}
}