Embracing Biased Transition Matrices for Complementary-Label Learning with Many Classes

Unknown Date

AI Summary

This work addresses the significant performance degradation of traditional complementary label learning as the number of classes increases, which limits its applicability to large-scale scenarios. To overcome this challenge, the authors propose the Bias-Induced Constrained Labeling (BICL) framework, which systematically uncovers and leverages a biased, non-uniform complementary label generation mechanism. By constraining complementary labels to subsets of classes, BICL enhances the learning signal through a combination of biased transition matrix modeling, a constrained label generation strategy, and end-to-end weakly supervised training. Experiments on CIFAR-100 and TinyImageNet-200 demonstrate over sevenfold accuracy improvements compared to existing methods, substantially advancing the feasibility of complementary label learning in large-scale real-world applications.

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. This limitation stems from the common assumption of uniform label generation in traditional methods, which fatally dilutes the learning signal in many-class settings. In this paper, we demonstrate that this long-standing barrier can be overcome by deliberately designing a biased (non-uniform) generation process that restricts complementary labels to a subset of classes. This finding motivates us to propose Bias-Induced Constrained Labeling (BICL), a principled framework spanning data collection to training that leverages this bias. BICL enables effective learning on CIFAR-100 and TinyImageNet-200, achieving more than sevenfold accuracy improvements over traditional methods. Our findings establish a new trajectory for making CLL feasible for many classes in real-world applications.

Problem

• complementary-label learning
• many-class classification
• label generation bias
• weakly supervised learning
• scalability

Innovation

• complementary-label learning
• biased transition matrix
• non-uniform label generation
• many-class classification
• weakly supervised learning