支持对抗样本防御的AI加速器架构设计

清风飞扬go

已于 2022-04-18 17:38:53 修改

阅读量511

点赞数

分类专栏： DNN加速器架构设计文章标签：深度学习神经网络机器学习硬件架构

于 2022-04-18 17:34:09 首次发布

版权

1 篇文章

订阅专栏

这篇博客探讨了对抗样本防御在AI加速器架构设计中的新兴趋势。文章提到了Dnnguard、Ptolemy和NASGuard等创新架构，它们旨在增强深度学习模型对对抗攻击的鲁棒性。2-in-1 Accelerator方案同时优化了对抗攻击的防御和效率，对于关注加速器领域的研究人员具有重要参考价值。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

看到一个综述文章，感觉支持对抗样本防御的AI加速器架构设计好像是个新方向呀。DNN加速器架构设计还能继续挖么。。。。。。。。找到几个相关的文献。。2-in-1 accelerator的文章对这个方向很看好，值得搞加速器的小伙伴关注下。

Wang X, Hou R, Zhao B, et al. Dnnguard: An elastic heterogeneous dnn accelerator architecture against adversarial attacks[C]//Proceedings of the Twenty-Fifth International Conference on Architectural Support for Programming Languages and Operating Systems. 2020: 19-34.
Gan Y, Qiu Y, Leng J, et al. Ptolemy: Architecture support for robust deep learning[C]//2020 53rd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO). IEEE, 2020: 241-255.
Wang X, Zhao B, Hou R, et al. NASGuard: a novel accelerator architecture for robust neural architecture search (NAS) networks[C]//2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (ISCA). IEEE, 2021: 776-789.
Fu Y, Zhao Y, Yu Q, et al. 2-in-1 Accelerator: Enabling Random Precision Switch for Winning Both Adversarial Robustness and Efficiency[C]//MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture. 2021: 225-237.