notesum.ai

Published at November 11

Adversarial Detection with a Dynamically Stable System

cs.AI

Released Date: November 11, 2024

Authors: Xiaowei Long¹, Jie Lin¹, Xiangyuan Yang¹

Aff.: ¹Xi'an Jiaotong University

Arxiv: http://arxiv.org/abs/2411.06666v1

Refer to caption

H	CIFAR10&VGG19							CIFAR10&ResNet50
	FGSM	PGD	CW	DeepFool	Square	AutoAttack		FGSM	PGD	CW	DeepFool	Square	AutoAttack
FS	0.7957_0.9675	0.9769_0.9978	0.9109_0.9755	0.9133_0.9612	0.8847_0.9726	0.9768_0.9939		0.8936_0.9938	0.9953_1.0000	0.8882_0.9841	0.8439_0.9636	0.7996_0.9617	0.9957_0.9999
KDBU	0.7570_0.9663	0.9789_0.9985	0.9235_0.9793	0.8514_0.9526	0.8724_0.9701	0.9433_0.9949		0.8719_0.9920	0.9946_0.9999	0.9480_0.9821	0.9663_0.9833	0.9683_0.9822	0.9918_0.9998
LID	0.9612_0.9820	0.9478_0.9980	0.9218_0.9764	0.8452_0.9577	0.9108_0.9770	0.9072_0.9973		0.9398_0.9943	0.9577_0.9999	0.8123_0.9828	0.7262_0.9624	0.7414_0.9642	0.9625_0.9998
MD	0.9458_0.9771	0.9954_0.9995	0.8700_0.9720	0.8051_0.9519	0.8555_0.9718	0.9504_0.9980		0.9936_0.9987	0.9509_0.9999	0.9134_0.9848	0.7970_0.9637	0.8763_0.9674	0.9790_0.9999
JTLA	0.7206_0.9671	0.5732_0.9944	0.8908_0.9802	0.8362_0.9605	0.9666_0.9866	0.5852_0.9908		0.8313_0.9931	0.6834_0.9999	0.9382_0.9912	0.9814_0.9897	0.9775_0.9868	0.7050_0.9998
LiBRe	0.7481_0.9670	0.6316_0.9939	0.8315_0.9728	0.8583_0.9570	0.8365_0.9721	0.6514_0.9910		0.7831_0.9919	0.5814_0.9999	0.8833_0.9844	0.9018_0.9668	0.9224_0.9660	0.5583_0.9998
EPS	0.7682_0.9662	0.7624_0.9939	0.9054_0.9886	0.5836_0.9518	0.5116_0.9711	0.8387_0.9955		0.6640_0.9919	0.8186_1.0000	0.7967_0.9846	0.5268_0.9599	0.5033_0.9592	0.9598_0.9999
DSS(ours)	0.9661	0.9939	0.9716	0.9517	0.9705	0.9910		0.9919	0.9999	0.9822	0.9597	0.9592	0.9998
H	CIFAR100&VGG19							CIFAR100&ResNet50
	FGSM	PGD	CW	DeepFool	Square	AutoAttack		FGSM	PGD	CW	DeepFool	Square	AutoAttack
FS	0.7559_0.9477	0.9669_0.9981	0.8091_0.9424	0.7961_0.8873	0.7744_0.9183	0.9703_0.9974		0.9521_0.9910	0.9902_0.9997	0.8222_0.9262	0.7102_0.8687	0.7443_0.9040	0.9919_0.9998
KDBU	0.8037_0.9479	0.9509_0.9977	0.8751_0.9425	0.8703_0.8822	0.7678_0.9166	0.9393_0.9961		0.9135_0.9846	0.9910_0.9995	0.8891_0.9434	0.8593_0.9031	0.8631_0.9158	0.9881_0.9995
LID	0.9419_0.9739	0.9338_0.9984	0.8403_0.9453	0.6989_0.8846	0.7581_0.9165	0.9395_0.9982		0.9151_0.9899	0.9369_0.9996	0.7042_0.9221	0.5896_0.8627	0.6974_0.9022	0.9377_0.9995
MD	0.9913_0.9924	0.9862_0.9987	0.9288_0.9608	0.8703_0.9148	0.9154_0.9464	0.9880_0.9985		0.9974_0.9993	0.9881_0.9998	0.8963_0.9464	0.7858_0.8914	0.8777_0.9301	0.9940_0.9993
JTLA	0.7904_0.9486	0.6545_0.9973	0.8889_0.9538	0.8794_0.9175	0.9149_0.9416	0.6538_0.9960		0.8307_0.9876	0.7066_0.9995	0.9029_0.9603	0.9344_0.9517	0.9052_0.9418	0.7446_0.9994
LiBRe	0.7976_0.9483	0.5374_0.9971	0.8022_0.9410	0.8449_0.8816	0.8028_0.9167	0.5338_0.9960		0.8058_0.9848	0.7385_0.9994	0.8265_0.9324	0.8190_0.8653	0.8135_0.8938	0.6657_0.9994
DSS(ours)	0.9469	0.9970	0.9408	0.8815	0.9159	0.9960		0.9846	0.9994	0.9202	0.8626	0.8921	0.9994