Systematic taxonomy synthesizing generative, contrastive, and predictive self-supervised learning algorithms across vision, natural language, multimodal representations, and graphs. Highlights sample efficiency bottlenecks, non-contrastive collapse mechanisms, and downstream cross-domain transfer bounds.
@article{zhang2025survey,
title = {A survey on self-supervised learning: Recent advances and open problems},
author = {Zhang, Jianping and Yang, Lei and Sajjadi Mohammadabadi, Seyed Mahmoud and Yan, Feng},
journal = {Neurocomputing},
volume = {655},
pages = {131409},
year = {2025},
publisher = {Elsevier},
doi = {10.1016/j.neucom.2024.131409}
}