Lost or Hidden? A Concept-Level Forgetting in Supervised Continual Learning

2026-05-10

AI Summary

This study addresses a critical limitation in continual learning research: the inability to distinguish whether conceptual information in model representations is genuinely lost or merely inaccessible. The work introduces, for the first time, a concept-level decomposition of forgetting by leveraging sparse autoencoders (SAEs) to construct a task-anchored latent space, framing forgetting along three dimensions—concept deletion, recoverability, and decodability. Through analysis of internal concept evolution in vision models, the authors demonstrate that most instances of “forgetting” stem not from irreversible information loss but from shifts in representational pathways that reduce decodability; under linear assumptions, the majority of concepts remain recoverable. This approach transcends conventional task-level evaluation paradigms and provides new insights into the fundamental mechanisms underlying forgetting in continual learning.

Abstract

Continual learning studies how models can adapt to new tasks while retaining previously acquired knowledge. Although a broad spectrum of methods has been proposed to mitigate catastrophic forgetting, the field remains predominantly performance-driven, with limited insight into what forgetting actually corresponds to within the vision model's representation space. Prior work has primarily analyzed forgetting through task-level performance or coarse measures of representational drift, without disentangling output-level accessibility from changes in finer-grained internal structure. To this end, we propose a diagnostic framework that leverages Sparse Autoencoders (SAEs) to define a task-anchored latent feature space, enabling analysis of how task-specific information evolves at a finer granularity, where individual SAE latents are treated as concept proxies for recurring and relatively disentangled visual patterns in the model's internal computations. Within this framework, we decompose forgetting into apparent concept deletion, recoverability, and decodability. We show that a large portion of seemingly lost concept-level information can often be recovered under linearity assumption, with concept decodability degrading as more tasks are introduced. Overall, our findings suggest that a significant part of concept-level forgetting can be attributed to changes in the representational accessibility rather than complete information erasure.

Problem

• continual learning
• catastrophic forgetting
• representation space
• concept-level forgetting
• latent features

Innovation

• Sparse Autoencoders
• Continual Learning
• Concept-level Forgetting
• Representational Accessibility
• Latent Feature Space