Embracing Biased Transition Matrices for Complementary-Label Learning with Many Classes
The paper introduces a new framework for complementary-label learning (CLL) that addresses the challenges of scaling to many classes. By employing a biased generation process for complementary labels, the authors propose Bias-Induced Constrained Labeling (BICL), which significantly improves learning outcomes. The framework demonstrates over sevenfold accuracy improvements on benchmark datasets, paving the way for practical applications in real-world scenarios.
- ▪Complementary-label learning (CLL) is a weakly supervised learning paradigm where instances are labeled with classes they do not belong to.
- ▪Traditional CLL methods struggle with large label spaces due to the assumption of uniform label generation.
- ▪The proposed Bias-Induced Constrained Labeling (BICL) framework restricts complementary labels to a subset of classes, enhancing learning efficiency.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.15586 (cs) [Submitted on 15 May 2026] Title:Embracing Biased Transition Matrices for Complementary-Label Learning with Many Classes Authors:Tan-Ha Mai, Chao-Kai Chiang, Han-Hwa Shih, Gang Niu, Masashi Sugiyama, Hsuan-Tien Lin View a PDF of the paper titled Embracing Biased Transition Matrices for Complementary-Label Learning with Many Classes, by Tan-Ha Mai and 5 other authors View PDF HTML (experimental) Abstract:Complementary-label learning (CLL) is a weakly supervised paradigm where instances are labeled with classes they do not belong to. Despite a decade of research, CLL methods remain competitive mainly on 10-class classification, with scaling to large label spaces continuing to be an enduring bottleneck.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
