Graceful Forgetting in Generative Language Models

2025-05-26

AI Summary

Pretraining-finetuning paradigms often suffer from negative transfer—where pretrained knowledge impedes downstream task performance—in generative language models (GLMs). To address this, we propose Learning With Forgetting (LWF), the first systematic integration of elegant forgetting into GLMs. LWF employs Fisher information-weighted parameter importance estimation to dynamically identify task-irrelevant knowledge, leveraging generation output confidence as a proxy for relevance. It then applies periodic, progressive parameter resetting to selectively discard such knowledge. Crucially, LWF resolves the incompatibility between existing forgetting methods and autoregressive architectures. Extensive experiments across diverse generative finetuning tasks—including abstractive summarization, dialogue generation, and code synthesis—demonstrate consistent and significant performance gains over strong baselines. Our results empirically validate that controlled, targeted forgetting enhances both task adaptability and learning plasticity in GLMs, offering a principled approach to mitigate negative transfer without architectural modification or additional inference overhead.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Recently, the pretrain-finetune paradigm has become a cornerstone in various deep learning areas. While in general the pre-trained model would promote both effectiveness and efficiency of downstream tasks fine-tuning, studies have shown that not all knowledge acquired during pre-training is beneficial. Some of the knowledge may actually bring detrimental effects to the fine-tuning tasks, which is also known as negative transfer. To address this problem, graceful forgetting has emerged as a promising approach. The core principle of graceful forgetting is to enhance the learning plasticity of the target task by selectively discarding irrelevant knowledge. However, this approach remains underexplored in the context of generative language models, and it is often challenging to migrate existing forgetting algorithms to these models due to architecture incompatibility. To bridge this gap, in this paper we propose a novel framework, Learning With Forgetting (LWF), to achieve graceful forgetting in generative language models. With Fisher Information Matrix weighting the intended parameter updates, LWF computes forgetting confidence to evaluate self-generated knowledge regarding the forgetting task, and consequently, knowledge with high confidence is periodically unlearned during fine-tuning. Our experiments demonstrate that, although thoroughly uncovering the mechanisms of knowledge interaction remains challenging in pre-trained language models, applying graceful forgetting can contribute to enhanced fine-tuning performance.

Problem

• Address negative transfer in generative language models
• Selectively discard irrelevant pre-trained knowledge
• Enhance fine-tuning performance via graceful forgetting

Innovation

• Selectively discards irrelevant pre-trained knowledge
• Uses Fisher Information Matrix for parameter updates
• Periodically unlearns high-confidence detrimental knowledge