Abstraction as a Memory-Efficient Inductive Bias for Continual Learning

2026-03-17

AI Summary

This work addresses catastrophic forgetting in online continual learning, a challenge exacerbated by interference from new knowledge and the high memory overhead of replay buffers commonly used in existing methods. To overcome these limitations, the authors propose Abstract-Augmented Training (AAT), which introduces structural abstraction as an inductive bias that requires no additional memory and entirely eliminates the need for replay mechanisms. AAT jointly optimizes concrete instances and their abstract representations—such as entity masks or proverbial distillations—within the loss function to encourage the model to capture latent relational structures shared across samples. Experiments demonstrate that, without any extra memory consumption, AAT matches or even surpasses strong experience replay baselines on both relational and narrative datasets, validating abstraction as an effective strategy for efficient continual learning.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Search and Optimization: Learning to Search
• Multiagent Systems: Multiagent Learning

Application Category

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

Abstract

The real world is non-stationary and infinitely complex, requiring intelligent agents to learn continually without the prohibitive cost of retraining from scratch. While online continual learning offers a framework for this setting, learning new information often interferes with previously acquired knowledge, causes forgetting and degraded generalization. To address this, we propose Abstraction-Augmented Training (AAT), a loss-level modification encouraging models to capture the latent relational structure shared across examples. By jointly optimizing over concrete instances and their abstract representations, AAT introduces a memory-efficient inductive bias that stabilizes learning in strictly online data streams, eliminating the need for a replay buffer. To capture the multi-faceted nature of abstraction, we introduce and evaluate AAT on two benchmarks: a controlled relational dataset where abstraction is realized through entity masking, and a narrative dataset where abstraction is expressed through shared proverbs. Our results show that AAT achieves performance comparable to or exceeding strong experience replay (ER) baselines, despite requiring zero additional memory and only minimal changes to the training objective. This work highlights structural abstraction as a powerful, memory-free alternative to ER.

Problem

• continual learning
• catastrophic forgetting
• online learning
• memory efficiency
• inductive bias

Innovation

• abstraction
• continual learning
• inductive bias
• memory-efficient
• online learning