Unlocking Compositional Generalization in Continual Few-Shot Learning

TL;DR

This paper introduces a novel framework that decouples representation learning from compositional inference using self-supervised Vision Transformers, enabling improved generalization to novel concepts in continual few-shot learning.

Contribution

It pioneers a dual-phase strategy leveraging ViT's geometry, achieving state-of-the-art generalization and minimal forgetting in continual learning.

Findings

Achieves state-of-the-art unseen-concept generalization.

Maintains minimal forgetting across benchmarks.

Utilizes a dual-phase approach with frozen backbone and holistic optimization.

Abstract

Object-centric representations promise a key property for few-shot learning: Rather than treating a scene as a single unit, a model can decompose it into individual object-level parts that can be matched and compared across different concepts. In practice, this potential is rarely realized. Continual learners either collapse scenes into global embeddings, or train with part-level matching objectives that tie representations too closely to seen patterns, leaving them unable to generalize to truly novel concepts. In this paper, we identify this fundamental structural conflict and pioneer a new paradigm that strictly decouples representation learning from compositional inference. Leveraging the inherent patch-level semantic geometry of self-supervised Vision Transformers (ViTs), our framework employs a dual-phase strategy. During training, slot representations are optimized entirely toward holistic class identity, preserving highly generalizable, object-level geometries. At inference, preserved slots are dynamically composed to match novel scenes. We demonstrate that this paradigm offers dual structural benefits: The frozen backbone naturally prevents representation drift, while our lightweight, holistic optimization preserves the features' capacity for novel-concept transfer. Extensive experiments validate this approach, achieving state-of-the-art unseen-concept generalization and minimal forgetting across standard continual learning benchmarks.