AdapterSwap: Continuous Training of LLMs with Data Removal and Access-Control Guarantees

2024-04-12

arXiv.org

AI Summary

To address the triple challenges of knowledge updating, access control, and verifiable forgetting faced by large language models (LLMs) in dynamic data environments, this paper proposes a modular knowledge management framework based on dynamic adapter swapping. The method integrates low-rank adaptation (LoRA), incremental training, access-policy embedding, and forgetting verification to enable runtime, on-demand composition and fine-grained governance of knowledge. Key contributions include: (i) the first support for millisecond-level, user-specific data view switching; (ii) document-level verifiable knowledge deletion—achieving near-zero recall post-deletion; and (iii) negligible performance degradation (<1.2% accuracy drop) on prior tasks, substantially mitigating catastrophic forgetting. Experiments demonstrate that the framework simultaneously preserves model stability, ensures data controllability and update efficiency, and satisfies privacy compliance requirements—establishing a novel paradigm for continual learning and trustworthy LLM deployment.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion

Application Category

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

Abstract

Large language models (LLMs) are increasingly capable of completing knowledge intensive tasks by recalling information from a static pretraining corpus. Here we are concerned with LLMs in the context of evolving data requirements. For instance: batches of new data that are introduced periodically; subsets of data with user-based access controls; or requirements on dynamic removal of documents with guarantees that associated knowledge cannot be recalled. We wish to satisfy these requirements while at the same time ensuring a model does not forget old information when new data becomes available. To address these issues, we introduce AdapterSwap, a training and inference scheme that organizes knowledge from a data collection into a set of low-rank adapters, which are dynamically composed during inference. Our experiments demonstrate AdapterSwap's ability to support efficient continual learning, while also enabling organizations to have fine-grained control over data access and deletion.

Problem

• continuous training of LLMs
• data removal guarantees
• access-control guarantees

Innovation

• Dynamic low-rank adapter composition
• Efficient continual learning support
• Fine-grained data access control