Recent advances in weakly supervised learning: New supervision paradigms, assumption relaxations, and practical solutions

2026-08-07

AI Summary

This work addresses the challenge of training deep learning models under realistic conditions where labels are often incomplete, imprecise, or inaccurate by proposing a unified algorithmic framework for weakly supervised learning. The framework encompasses three representative settings: confidence-based disambiguation, learning from complementary labels, and partial label learning. It introduces a novel paradigm for confidence-based disambiguation, relaxes restrictive data generation assumptions in complementary label learning, and establishes a fair evaluation protocol for partial label learning. Grounded in consistency theory and standardized modeling principles, the proposed approach significantly enhances model robustness and accuracy across diverse weak supervision scenarios while advancing standardization and reproducibility in evaluation protocols within the field.

Abstract

Deep learning has achieved great success in recent years thanks to the availability of high-quality, well-annotated training data. However, this requirement is often not met in real-world applications. Weakly supervised learning aims to train an accurate model with incomplete, inexact, or inaccurate supervision. In this chapter, we will discuss recent advances in this field, including new supervision paradigms, relaxed assumptions, and practical solutions. First, we introduce a new weakly supervised binary classification problem called confidence-difference classification and propose consistent approaches to solve it. Next, we investigate complementary-label learning, a weakly supervised multi-class classification problem. Our proposed approaches are based on more relaxed assumptions about the data generation process than existing consistent approaches. Lastly, we present an evaluation framework for partial-label learning, another popular multi-class weakly supervised learning problem, in order to promote fair and realistic evaluation of algorithms in this field.

Problem

• weakly supervised learning
• incomplete supervision
• inexact supervision
• inaccurate supervision
• real-world applications

Innovation

• confidence-difference classification
• complementary-label learning
• partial-label learning
• weakly supervised learning
• relaxed assumptions