English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Data-efficient learning has garnered significant attention, especially giventhe current trend of large multi-modal models. Recently, dataset distillationbecomes an effective approach for data-efficiency; however, the distillationprocess itself can still be inefficient. In this work, we model the datasetdistillation task within the context of information transport. By observing thesubstantial data redundancy inherent in the distillation, we argue to put moreemphasis on the samples' utility for the distillation task. We introduce andvalidate a family of data utility estimators and optimal data selection methodsto exploit the most valuable samples. This strategy significantly reduces thetraining costs and extends various existing distillation algorithms to largerand more diversified datasets, e.g., in some cases only 0.04% training data issufficient for comparable distillation performance. Our method consistentlyenhances the distillation algorithms, even on much larger-scale and moreheterogeneous datasets, e.g. ImageNet-1K and Kinetics-400. This paradigm opensup new avenues in the dynamics of distillation and paves the way for efficientdataset distillation. Our code is available onhttps://github.com/silicx/GoldFromOres .

Distill Gold from Massive Ores: Efficient Dataset Distillation via Critical Samples Selection