Semantic Redundancies in Image-Classification Datasets: The 10% You Don't Need
This addresses data efficiency for machine learning practitioners by revealing redundancies in common benchmarks, though it is incremental as it builds on existing dataset analysis.
The paper tackled the problem of semantic redundancies in image-classification datasets by identifying subsets that generalize as well as the full dataset, finding at least 10% redundancy in CIFAR-10 and ImageNet.
Large datasets have been crucial to the success of deep learning models in the recent years, which keep performing better as they are trained with more labelled data. While there have been sustained efforts to make these models more data-efficient, the potential benefit of understanding the data itself, is largely untapped. Specifically, focusing on object recognition tasks, we wonder if for common benchmark datasets we can do better than random subsets of the data and find a subset that can generalize on par with the full dataset when trained on. To our knowledge, this is the first result that can find notable redundancies in CIFAR-10 and ImageNet datasets (at least 10%). Interestingly, we observe semantic correlations between required and redundant images. We hope that our findings can motivate further research into identifying additional redundancies and exploiting them for more efficient training or data-collection.