paper-with-me

홈 › Papers

Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs

2024-06-14 · Abhimanyu Hans, Yuxin Wen, Neel Jain, John Kirchenbauer, Hamid Kazemi, Prajwal Singhania, Siddharth Singh, Gowthami Somepalli, Jonas Geiping, Abhinav Bhatele, Tom Goldstein

Large language models can memorize and repeat their training data, causing privacy and copyright risks. To mitigate memorization, we introduce a subtle modification to the next-token training objective that we call the goldfish loss. During training, randomly sampled subsets of tokens are excluded from the loss computation. These dropped tokens are not memorized by the model, which prevents verbatim reproduction of a complete chain of tokens from the training set. We run extensive experiments training billion-scale Llama-2 models, both pre-trained and trained from scratch, and demonstrate significant reductions in extractable memorization with little to no impact on downstream benchmarks.

📄 PDF Abstract BibTeX arXiv:2406.10209

Code (1)

ahans30/goldfish-loss 공식 구현 pytorch

Tasks

Memorization

Similar Papers 제목 키워드 기반

Unconsciously Forget: Mitigating Memorization; Without Knowing What is being Memorized

2025-12-10 · Er Jin, Yang Zhang, Yongli Mou, Yanfei Dong 외 arxiv

Recent advances in generative models have demonstrated an exceptional ability to produce highly realistic images. However, previous studies show that generated images often resemble the training data, and this problem be…

Localizing and Mitigating Memorization in Image Autoregressive Models

2025-08-30 · Aditya Kasliwal, Franziska Boenisch, Adam Dziedzic arxiv

Image AutoRegressive (IAR) models have achieved state-of-the-art performance in speed and quality of generated images. However, they also raise concerns about memorization of their training data and its implications for …

Detecting and Mitigating Memorization in Diffusion Models through Anisotropy of the Log-Probability

2026-01-28 · Rohan Asthana, Vasileios Belagiannis arxiv

Diffusion-based image generative models produce high-fidelity images through iterative denoising but remain vulnerable to memorization, where they unintentionally reproduce exact copies or parts of training images. Recen…

Mitigating Unintended Memorization with LoRA in Federated Learning for LLMs

2025-02-07 · Thierry Bossy, Julien Vignoud, Tahseen Rabbani, Juan R. Troncoso Pastoriza 외

Federated learning (FL) is a popular paradigm for collaborative training which avoids direct data exposure between clients. However, data privacy issues still remain: FL-trained large language models are capable of memor…

Federated LearningMedical Question AnsweringMemorizationPrivacy Preserving+1

A Geometric Framework for Understanding Memorization in Generative Models

2024-10-31 · Brendan Leigh Ross, Hamidreza Kamkari, Tongzi Wu, Rasa Hosseinzadeh 외

As deep generative models have progressed, recent work has shown them to be capable of memorizing and reproducing training datapoints when deployed. These findings call into question the usability of generative models, e…

Memorization