28 August 2023
Byung Hyun Lee,
Se Young Chun
Proposes hierarchical label expansion (HLE), a new continual learning setup mirroring real-world knowledge accumulation
Introduces pseudo-labeling based memory management and flexible memory sampling for HLE
Demonstrates significant performance gains on HLE over prior CL methods
Shows strong results on HLE while still outperforming on existing CL setups
Provides comprehensive analysis and ablation studies validating contributions
Online learning of image classifiers with expanding labels
This paper proposes a new continual learning setup called hierarchical label expansion that simulates real-world knowledge accumulation, where learning begins with coarse categories that expand into finer subcategories over time. The paper introduces a method to tackle this setup using pseudo-labeling, flexible memory sampling, and hierarchy-aware training.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper