Towards Robust Incremental Learning under Ambiguous Supervision

TL;DR

This paper introduces a new weakly-supervised incremental learning framework called IPLL, which effectively handles label ambiguity and reduces forgetting using prototype-guided disambiguation and replay techniques.

Contribution

The paper proposes IPLL and PGDR, novel methods for robust incremental learning under ambiguous supervision, addressing label ambiguity and catastrophic forgetting simultaneously.

Findings

PGDR achieves superior disambiguation of samples.

The framework reduces catastrophic forgetting effectively.

Experiments demonstrate improved performance over existing methods.

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