AI4Science researchers, Marjan Stoimchev, Jurica Levatić, Dragi Kocev, and Sašo Džeroski, published SSL-MAE, an adaptive semi-supervised learning framework for multi-label classification of remote sensing images.

Built on masked autoencoders, SSL-MAE unifies self-supervised and discriminative learning to efficiently leverage abundant unlabeled data alongside scarce labeled examples. The method introduces an adaptive mechanism to balance supervision and learning, significantly improving performance across diverse remote sensing image datasets.

Comprehensive experiments on 10 public datasets demonstrate that SSL-MAE outperforms current state-of-the-art semi-supervised approaches and enhances the predictive capabilities of various self-supervised methods. Furthermore, the proposed adaptive joint learning strategy is not restricted to a single design choice; instead, when integrated with different state-of-the-art self-supervised approaches, it significantly enhances their predictive performance.

For more information, visit: https://ieeexplore.ieee.org/document/11029124