GeoEdit: Geometric Knowledge Editing for Large Language Models

TL;DR

GeoEdit is a novel framework that improves knowledge editing in large language models by leveraging geometric relationships of parameter updates to better incorporate new knowledge while preserving existing knowledge.

Contribution

GeoEdit introduces a direction-aware, geometry-based approach for more effective and precise knowledge editing in large language models, addressing limitations of previous training-based methods.

Findings

Outperforms existing state-of-the-art methods in experiments

Effectively preserves general knowledge while updating specific facts

Utilizes geometric relationships for more accurate knowledge editing

Abstract

Regular updates are essential for maintaining up-to-date knowledge in large language models (LLMs). Consequently, various model editing methods have been developed to update specific knowledge within LLMs. However, training-based approaches often struggle to effectively incorporate new knowledge while preserving unrelated general knowledge. To address this challenge, we propose a novel framework called Geometric Knowledge Editing (GeoEdit). GeoEdit utilizes the geometric relationships of parameter updates from fine-tuning to differentiate between neurons associated with new knowledge updates and those related to general knowledge perturbations. By employing a direction-aware knowledge identification method, we avoid updating neurons with directions approximately orthogonal to existing knowledge, thus preserving the model's generalization ability. For the remaining neurons, we integrate both old and new knowledge for aligned directions and apply a "forget-then-learn" editing strategy for opposite directions. Additionally, we introduce an importance-guided task vector fusion technique that filters out redundant information and provides adaptive neuron-level weighting, further enhancing model editing performance. Extensive experiments on two publicly available datasets demonstrate the superiority of GeoEdit over existing state-of-the-art methods.