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

Label-Efficient Dataset Pruning via Semi-Supervised Pseudo-Labeling

About

Dataset pruning reduces the storage and training costs of deep learning by selecting an informative subset from a large dataset. However, most existing pruning methods require fully labeled data, which limits their applicability in realistic settings where unlabeled data are abundant and annotation is costly. Recent label-free pruning methods address this issue, but they rely on features from pretrained models to estimate example difficulty. This dependence can be unreliable when the target dataset differs substantially from the pretraining distribution. We propose SemiPrune, a label-efficient dataset pruning framework, using only a small randomly labeled subset, that uses semi-supervised learning to generate pseudo-labels for unlabeled data, allowing existing supervised pruning methods that require label information to be seamlessly applied to the resulting pseudo-labeled training pool. We then estimate example difficulty from pseudo-label-induced training dynamics and select a coreset. By learning directly from the target dataset, our method better captures the target distribution and provides more reliable signals for difficulty estimation and coreset selection. We validate our approach on domain-specific, image-corrupted, and long-tailed datasets, where it achieves state-of-the-art performance among label-free and label-efficient baselines, while also demonstrating competitive performance on standard benchmarks.

Yeseul Cho, Baekrok Shin, Changmin Kang, Chulhee Yun• 2026

Related benchmarks

TaskDatasetResultRank
Image ClassificationSUN397 (test)
Top-1 Accuracy60.7
231
Image ClassificationFood-101 (test)
Accuracy77.9
145
Image ClassificationCIFAR-100-LT balanced imbalance factor 0.1 (test)
Accuracy58.3
45
Image ClassificationCIFAR-100 LT IF=0.01 (test)
Accuracy38
45
Image ClassificationCIFAR-100-C 30% corrupted (test)
Accuracy73.3
45
Image ClassificationTiny-ImageNet-C 30% corrupted (test)
Accuracy42
45
Image ClassificationCaltech-101 naturally imbalanced (test)
Accuracy79
45
Image ClassificationCIFAR-100 (test)
Accuracy (k=30)77.4
12
Image ClassificationCIFAR-10 (test)
Accuracy (30%)95.4
9
Image ClassificationImageNet 1k (test)
Accuracy (30% Threshold)73.3
9
Showing 10 of 10 rows

Other info

Follow for update