Improving Unsupervised Image Clustering With Robust Learning
This work provides an incremental improvement for existing unsupervised image clustering methods by enhancing their robustness and calibration.
This paper addresses the issue of faulty predictions and overconfidence in unsupervised image clustering by proposing RUC, a model inspired by robust learning. RUC uses pseudo-labels from existing clustering models as a noisy dataset to revise misaligned knowledge and reduce overconfidence, leading to better calibration and robustness against adversarial noise.
Unsupervised image clustering methods often introduce alternative objectives to indirectly train the model and are subject to faulty predictions and overconfident results. To overcome these challenges, the current research proposes an innovative model RUC that is inspired by robust learning. RUC's novelty is at utilizing pseudo-labels of existing image clustering models as a noisy dataset that may include misclassified samples. Its retraining process can revise misaligned knowledge and alleviate the overconfidence problem in predictions. The model's flexible structure makes it possible to be used as an add-on module to other clustering methods and helps them achieve better performance on multiple datasets. Extensive experiments show that the proposed model can adjust the model confidence with better calibration and gain additional robustness against adversarial noise.