Adversarial Pruning Benchmark
收藏资源简介:
Adversarial Pruning Benchmark是由卡利亚里大学创建的一个用于评估对抗性剪枝方法的数据集。该数据集包含26种不同的对抗性剪枝方法,旨在通过剪枝技术减少神经网络的大小同时保持其对抗攻击的鲁棒性。数据集的内容涵盖了不同的剪枝策略和训练阶段的应用,旨在解决在资源受限场景下模型压缩与对抗鲁棒性之间的平衡问题。数据集的创建过程涉及对现有方法的分类和评估,最终形成了一个统一的评估基准,以促进对抗性剪枝方法的研究和发展。
Adversarial Pruning Benchmark is a dataset for evaluating adversarial pruning methods, created by the University of Cagliari. This dataset encompasses 26 distinct adversarial pruning methods, aiming to reduce the size of neural networks via pruning techniques while maintaining their robustness against adversarial attacks. The dataset covers applications of different pruning strategies across various training stages, aiming to address the trade-off between model compression and adversarial robustness in resource-constrained scenarios. The creation of this dataset involves the classification and evaluation of existing methods, ultimately forming a unified evaluation benchmark to facilitate the research and advancement of adversarial pruning methods.

- 1Adversarial Pruning: A Survey and Benchmark of Pruning Methods for Adversarial Robustness卡利亚里大学 · 2024年



