Towards Robust Incremental Learning under Ambiguous Supervision

2025-01-23

AI Summary

To address class ambiguity and catastrophic forgetting caused by label ambiguity and high annotation costs in dynamic scenarios, this paper proposes Incremental Partial Label Learning (IPLL), a novel paradigm that tightly integrates partial label learning with incremental learning for the first time. To jointly tackle the coupled challenges of label uncertainty and knowledge forgetting, we design Prototype-Guided Disambiguation and Replay (PGDR): it achieves accurate disambiguation via prototype-driven pseudo-label initialization and momentum-based optimization; and mitigates forgetting through representative-diverse memory sampling coupled with knowledge distillation-based replay. Extensive experiments on multiple benchmarks demonstrate that IPLL significantly improves disambiguation accuracy on new tasks (+8.2%–14.6%) and substantially alleviates performance degradation on old classes (reducing forgetting rate by 37.5%). These results validate IPLL’s effectiveness and robustness in weakly supervised incremental learning settings.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Search and Optimization: Learning to Search
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Traditional Incremental Learning (IL) targets to handle sequential fully-supervised learning problems where novel classes emerge from time to time. However, due to inherent annotation uncertainty and ambiguity, collecting high-quality annotated data in a dynamic learning system can be extremely expensive. To mitigate this problem, we propose a novel weakly-supervised learning paradigm called Incremental Partial Label Learning (IPLL), where the sequentially arrived data relate to a set of candidate labels rather than the ground truth. Technically, we develop the Prototype-Guided Disambiguation and Replay Algorithm (PGDR) which leverages the class prototypes as a proxy to mitigate two intertwined challenges in IPLL, i.e., label ambiguity and catastrophic forgetting. To handle the former, PGDR encapsulates a momentum-based pseudo-labeling algorithm along with prototype-guided initialization, resulting in a balanced perception of classes. To alleviate forgetting, we develop a memory replay technique that collects well-disambiguated samples while maintaining representativeness and diversity. By jointly distilling knowledge from curated memory data, our framework exhibits a great disambiguation ability for samples of new tasks and achieves less forgetting of knowledge. Extensive experiments demonstrate that PGDR achieves superior

Problem

• Incremental Learning
• Class Ambiguity
• Catastrophic Forgetting

Innovation

• Incremental Partial Label Learning
• Prototype Guided Deblur and Replay (PGDR) Algorithm
• Class Ambiguity Resolution