Pruning as a Defense: Reducing Memorization in Large Language Models

Large language models have been shown to memorize significant portions of their training data, which they can reproduce when appropriately prompted. This work investigates the impact of simple pruning techniques on this behavior. Our findings reveal that pruning effectively reduces the extent of memorization in LLMs, demonstrating its potential as a foundational approach for mitigating membership inference attacks.

@article{gupta2025_2502.15796,
  title={ Pruning as a Defense: Reducing Memorization in Large Language Models },
  author={ Mansi Gupta and Nikhar Waghela and Sarthak Gupta and Shourya Goel and Sanjif Shanmugavelu },
  journal={arXiv preprint arXiv:2502.15796},
  year={ 2025 }
}