Exploring Criteria of Loss Reweighting to Enhance LLM Unlearning

• URL: http://arxiv.org/abs/2505.11953v2
• Date: Wed, 28 May 2025 03:33:24 GMT
• Title: Exploring Criteria of Loss Reweighting to Enhance LLM Unlearning
• Authors: Puning Yang, Qizhou Wang, Zhuo Huang, Tongliang Liu, Chengqi Zhang, Bo Han,
• Abstract summary: Loss reweighting has shown significant benefits for machine unlearning with large language models (LLMs)<n>In this paper, we identify two distinct goals of loss reweighting, namely, Saturation and Importance.<n>We propose SatImp, a simple reweighting method that combines the advantages of both saturation and importance.
• Score: 66.8042627609456
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Loss reweighting has shown significant benefits for machine unlearning with large language models (LLMs). However, their exact functionalities are left unclear and the optimal strategy remains an open question, thus impeding the understanding and improvement of existing methodologies. In this paper, we identify two distinct goals of loss reweighting, namely, Saturation and Importance -- the former indicates that those insufficiently optimized data should be emphasized, while the latter stresses some critical data that are most influential for loss minimization. To study their usefulness, we design specific reweighting strategies for each goal and evaluate their respective effects on unlearning. We conduct extensive empirical analyses on well-established benchmarks, and summarize some important observations as follows: (i) Saturation enhances efficacy more than importance-based reweighting, and their combination can yield additional improvements. (ii) Saturation typically allocates lower weights to data with lower likelihoods, whereas importance-based reweighting does the opposite. (iii) The efficacy of unlearning is also largely influenced by the smoothness and granularity of the weight distributions. Based on these findings, we propose SatImp, a simple reweighting method that combines the advantages of both saturation and importance. Empirical results on extensive datasets validate the efficacy of our method, potentially bridging existing research gaps and indicating directions for future research. Our code is available at https://github.com/tmlr-group/SatImp.

Related papers

• Understand the Effect of Importance Weighting in Deep Learning on Dataset Shift [0.0]
  We evaluate the effectiveness of importance weighting in deep neural networks under label shift and covariate shift.<n>We observe that weighting strongly affects decision boundaries early in training but fades with prolonged optimization.<n>Our results call into question the practical utility of importance weighting for real-world distribution shifts.
  arXiv  Detail & Related papers  (2025-05-06T15:16:38Z)
• Dynamic Loss-Based Sample Reweighting for Improved Large Language Model Pretraining [55.262510814326035]
  Existing reweighting strategies primarily focus on group-level data importance.<n>We introduce novel algorithms for dynamic, instance-level data reweighting.<n>Our framework allows us to devise reweighting strategies deprioritizing redundant or uninformative data.
  arXiv  Detail & Related papers  (2025-02-10T17:57:15Z)
• SEEKR: Selective Attention-Guided Knowledge Retention for Continual Learning of Large Language Models [27.522743690956315]
  We propose a SElective attEntion-guided Knowledge Retention method (SEEKR) for data-efficient replay-based continual learning of large language models (LLMs) SEEKR performs attention distillation on the selected attention heads for finer-grained knowledge retention. Experimental results on two continual learning benchmarks for LLMs demonstrate the superiority of SEEKR over the existing methods on both performance and efficiency.
  arXiv  Detail & Related papers  (2024-11-09T13:02:36Z)
• TIS-DPO: Token-level Importance Sampling for Direct Preference Optimization With Estimated Weights [73.9088920210495]
  We propose a token-level importance sampling DPO objective named TIS-DPO that assigns importance weights to each token based on its reward.<n>TIS-DPO significantly outperforms various baseline methods on harmlessness and helpfulness alignment and summarization tasks.
  arXiv  Detail & Related papers  (2024-10-06T04:03:00Z)
• Towards Effective Evaluations and Comparisons for LLM Unlearning Methods [97.2995389188179]
  This paper seeks to refine the evaluation of machine unlearning for large language models.<n>It addresses two key challenges -- the robustness of evaluation metrics and the trade-offs between competing goals.
  arXiv  Detail & Related papers  (2024-06-13T14:41:00Z)
• Rethinking Deep Contrastive Learning with Embedding Memory [58.66613563148031]
  Pair-wise loss functions have been extensively studied and shown to continuously improve the performance of deep metric learning (DML) We provide a new methodology for systematically studying weighting strategies of various pair-wise loss functions, and rethink pair weighting with an embedding memory.
  arXiv  Detail & Related papers  (2021-03-25T17:39:34Z)
• Semi-Supervised Off Policy Reinforcement Learning [3.48396189165489]
  Health-outcome information is often not well coded but rather embedded in clinical notes. We propose a semi-supervised learning (SSL) approach that efficiently leverages a small sized labeled data with true outcome observed, and a large unlabeled data with outcome surrogates. Our method is at least as efficient as the supervised approach, and moreover safe as it robust to mis-specification of the imputation models.
  arXiv  Detail & Related papers  (2020-12-09T00:59:12Z)
• ReMP: Rectified Metric Propagation for Few-Shot Learning [67.96021109377809]
  A rectified metric space is learned to maintain the metric consistency from training to testing. Numerous analyses indicate that a simple modification of the objective can yield substantial performance gains. The proposed ReMP is effective and efficient, and outperforms the state of the arts on various standard few-shot learning datasets.
  arXiv  Detail & Related papers  (2020-12-02T00:07:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.