FGGM: Fisher-Guided Gradient Masking for Continual Learning

TL;DR

FGGM is a novel continual learning framework that uses Fisher-guided gradient masking to selectively preserve important parameters, significantly reducing catastrophic forgetting in large language models without needing past data.

Contribution

It introduces Fisher-Guided Gradient Masking, a mathematically principled method for parameter importance estimation that improves continual learning performance.

Findings

9.6% relative improvement on TRACE benchmark

4.4% improvement over MIGU on TRACE tasks

Reduced forgetting in code generation tasks

Abstract

Catastrophic forgetting impairs the continuous learning of large language models. We propose Fisher-Guided Gradient Masking (FGGM), a framework that mitigates this by strategically selecting parameters for updates using diagonal Fisher Information. FGGM dynamically generates binary masks with adaptive thresholds, preserving critical parameters to balance stability and plasticity without requiring historical data. Unlike magnitude-based methods such as MIGU, our approach offers a mathematically principled parameter importance estimation. On the TRACE benchmark, FGGM shows a 9.6% relative improvement in retaining general capabilities over supervised fine-tuning (SFT) and a 4.4% improvement over MIGU on TRACE tasks. Additional analysis on code generation tasks confirms FGGM's superior performance and reduced forgetting, establishing it as an effective solution.