CollabEdit: Towards Non-destructive Collaborative Knowledge Editing
This work addresses the need for efficient and private collaborative editing of LLM knowledge, which is incremental as it extends existing knowledge editing methods to a multi-party setting.
The paper tackles the problem of collaborative knowledge editing for large language models, where multiple parties edit knowledge in a privacy-preserving and continual manner, by proposing the COLLABEDIT framework that addresses challenges like knowledge overlap, conflict, and forgetting, and demonstrates superiority over destructive baselines in experiments on two datasets.
Collaborative learning of large language models (LLMs) has emerged as a new paradigm for utilizing private data from different parties to guarantee efficiency and privacy. Meanwhile, Knowledge Editing (KE) for LLMs has also garnered increased attention due to its ability to manipulate the behaviors of LLMs explicitly, yet leaves the collaborative KE case (in which knowledge edits of multiple parties are aggregated in a privacy-preserving and continual manner) unexamined. To this end, this manuscript dives into the first investigation of collaborative KE, in which we start by carefully identifying the unique three challenges therein, including knowledge overlap, knowledge conflict, and knowledge forgetting. We then propose a non-destructive collaborative KE framework, COLLABEDIT, which employs a novel model merging mechanism to mimic the global KE behavior while preventing the severe performance drop. Extensive experiments on two canonical datasets demonstrate the superiority of COLLABEDIT compared to other destructive baselines, and results shed light on addressing three collaborative KE challenges and future applications. Our code is available at https://github.com/LINs-lab/CollabEdit.