FGGM: Fisher-Guided Gradient Masking for Continual Learning
This addresses the problem of maintaining model capabilities during continual learning for AI practitioners, offering an incremental improvement over existing methods.
The paper tackles catastrophic forgetting in large language models by proposing Fisher-Guided Gradient Masking (FGGM), which uses diagonal Fisher Information to select parameters for updates, resulting in a 9.6% relative improvement over supervised fine-tuning and a 4.4% improvement over MIGU on the TRACE benchmark.
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.