Dataset Distillation by Influence Matching
Computer Vision and Pattern Recognition
Summary
The authors propose a new way to create small synthetic datasets that capture the important effects of full large datasets on model training outcomes. Instead of matching the step-by-step training process, their method called Influence Matching focuses on matching the final impact on model parameters. They introduce a fast, differentiable way to estimate how adding or removing data shifts model parameters and use this to optimize the synthetic data. Their method performs better than previous techniques on image classification and vision-language tasks.
Authors
Haoru Tan, Wang Wang, Sitong Wu, Xiuzhe Wu, Yangtian Sun, Chirui Chang, Shaofeng Zhang, Xiaojuan Qi
Abstract
We revisit dataset distillation from an outcome-centric perspective. Rather than aligning process surrogates (per-step gradients or training trajectories), Influence Matching (Inf-Match) aligns the final outcome of training: it learns a compact synthetic set whose effect on the converged parameters matches that of the full dataset. Concretely, we introduce a fully differentiable, sample-level influence estimator that quantifies parameter shifts from adding or removing data, without time-consuming inverse-Hessian products or convexity assumptions. The estimator runs in linear time by unrolling the optimization dynamics and applying a first-order Taylor approximation. We then learn the synthetic set by minimizing the mismatch between its influence and that of the real dataset, yielding outcome alignment rather than heuristic process imitation. Inf-Match delivers the best accuracy across standard classification benchmarks. For instance, on Tiny-ImageNet (IPC=10), Inf-Match attains 31.5\%, a +4.7\% improvement over NCFM. Beyond classification, Inf-Match scales to vision-language distillation on Flickr30K, outperforming strong process-matching baselines. For instance, with 200 to 1000 synthetic samples, our method achieved a leading impressive average on image/text retrieval tasks, higher than NCFM by 2.5\%. The code will be released via https://github.com/hrtan/infmatch.