RECALL: REpresentation-aligned Catastrophic-forgetting ALLeviation via Hierarchical Model Merging

2025-10-23

AI Summary

To address severe catastrophic forgetting and reliance on historical data in continual learning (CL) for large language models (LLMs), this paper proposes RECALL—a data-free, hierarchical adaptive CL framework. RECALL innovatively employs inter-layer hidden representations as knowledge proxies: it clusters representative samples to compute cross-task representation similarity and performs hierarchical model merging—preserving generic features in shallow layers while integrating task-specific knowledge in deeper layers—without requiring task labels or stored historical data. Evaluated across five NLP tasks and diverse CL settings, RECALL consistently outperforms state-of-the-art baselines, achieving superior knowledge retention and new-task generalization without performance trade-offs. This work establishes a novel paradigm for efficient, scalable online evolution of LLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

We unveil that internal representations in large language models (LLMs) serve as reliable proxies of learned knowledge, and propose RECALL, a novel representation-aware model merging framework for continual learning without access to historical data. RECALL computes inter-model similarity from layer-wise hidden representations over clustered typical samples, and performs adaptive, hierarchical parameter fusion to align knowledge across models. This design enables the preservation of domain-general features in shallow layers while allowing task-specific adaptation in deeper layers. Unlike prior methods that require task labels or incur performance trade-offs, RECALL achieves seamless multi-domain integration and strong resistance to catastrophic forgetting. Extensive experiments across five NLP tasks and multiple continual learning scenarios show that RECALL outperforms baselines in both knowledge retention and generalization, providing a scalable and data-free solution for evolving LLMs.

Problem

• Prevents catastrophic forgetting in continual learning without historical data
• Aligns knowledge across models via hierarchical parameter fusion
• Preserves domain-general features while enabling task-specific adaptation

Innovation

• Hierarchical model merging for continual learning
• Layer-wise parameter fusion using representation similarity
• Data-free domain integration without performance trade-offs