Matches in SemOpenAlex for { <https://semopenalex.org/work/W3206536361> ?p ?o ?g. }
- W3206536361 abstract "Recent work argues that robust training requires substantially larger datasets than those required for standard classification. On CIFAR-10 and CIFAR-100, this translates into a sizable robust-accuracy gap between models trained solely on data from the original training set and those trained with additional data extracted from the 80 Million Tiny Images dataset (TI-80M). In this paper, we explore how generative models trained solely on the original training set can be leveraged to artificially increase the size of the original training set and improve adversarial robustness to $ell_p$ norm-bounded perturbations. We identify the sufficient conditions under which incorporating additional generated data can improve robustness, and demonstrate that it is possible to significantly reduce the robust-accuracy gap to models trained with additional real data. Surprisingly, we even show that even the addition of non-realistic random data (generated by Gaussian sampling) can improve robustness. We evaluate our approach on CIFAR-10, CIFAR-100, SVHN and TinyImageNet against $ell_infty$ and $ell_2$ norm-bounded perturbations of size $epsilon = 8/255$ and $epsilon = 128/255$, respectively. We show large absolute improvements in robust accuracy compared to previous state-of-the-art methods. Against $ell_infty$ norm-bounded perturbations of size $epsilon = 8/255$, our models achieve 66.10% and 33.49% robust accuracy on CIFAR-10 and CIFAR-100, respectively (improving upon the state-of-the-art by +8.96% and +3.29%). Against $ell_2$ norm-bounded perturbations of size $epsilon = 128/255$, our model achieves 78.31% on CIFAR-10 (+3.81%). These results beat most prior works that use external data." @default.
- W3206536361 created "2021-10-25" @default.
- W3206536361 creator A5006380946 @default.
- W3206536361 creator A5029801689 @default.
- W3206536361 creator A5046296629 @default.
- W3206536361 creator A5051904433 @default.
- W3206536361 creator A5059632022 @default.
- W3206536361 creator A5061287315 @default.
- W3206536361 date "2021-10-18" @default.
- W3206536361 modified "2023-10-01" @default.
- W3206536361 title "Improving Robustness using Generated Data" @default.
- W3206536361 cites W1988720110 @default.
- W3206536361 cites W2145607950 @default.
- W3206536361 cites W2194775991 @default.
- W3206536361 cites W2335728318 @default.
- W3206536361 cites W2342840547 @default.
- W3206536361 cites W2462831000 @default.
- W3206536361 cites W2512971201 @default.
- W3206536361 cites W2746314669 @default.
- W3206536361 cites W2774644650 @default.
- W3206536361 cites W2783879794 @default.
- W3206536361 cites W2786104118 @default.
- W3206536361 cites W2804047946 @default.
- W3206536361 cites W2886281300 @default.
- W3206536361 cites W2891229686 @default.
- W3206536361 cites W2892179671 @default.
- W3206536361 cites W2893749619 @default.
- W3206536361 cites W2897355816 @default.
- W3206536361 cites W2898152545 @default.
- W3206536361 cites W2899663614 @default.
- W3206536361 cites W2945027741 @default.
- W3206536361 cites W2947294642 @default.
- W3206536361 cites W2951165433 @default.
- W3206536361 cites W2962729158 @default.
- W3206536361 cites W2962872506 @default.
- W3206536361 cites W2962972504 @default.
- W3206536361 cites W2963060032 @default.
- W3206536361 cites W2963143631 @default.
- W3206536361 cites W2963173418 @default.
- W3206536361 cites W2963207607 @default.
- W3206536361 cites W2963263347 @default.
- W3206536361 cites W2963399829 @default.
- W3206536361 cites W2963542245 @default.
- W3206536361 cites W2963557656 @default.
- W3206536361 cites W2963564844 @default.
- W3206536361 cites W2963744840 @default.
- W3206536361 cites W2963857521 @default.
- W3206536361 cites W2964121744 @default.
- W3206536361 cites W2964137095 @default.
- W3206536361 cites W2964153729 @default.
- W3206536361 cites W2964197269 @default.
- W3206536361 cites W2964253222 @default.
- W3206536361 cites W2970203973 @default.
- W3206536361 cites W2970680991 @default.
- W3206536361 cites W2970780203 @default.
- W3206536361 cites W2970864518 @default.
- W3206536361 cites W2971316968 @default.
- W3206536361 cites W2974008169 @default.
- W3206536361 cites W2981016037 @default.
- W3206536361 cites W2992308087 @default.
- W3206536361 cites W2992535490 @default.
- W3206536361 cites W2995801453 @default.
- W3206536361 cites W2996582215 @default.
- W3206536361 cites W3006044868 @default.
- W3206536361 cites W3007305010 @default.
- W3206536361 cites W3009542902 @default.
- W3206536361 cites W3014852036 @default.
- W3206536361 cites W3034339621 @default.
- W3206536361 cites W3035032188 @default.
- W3206536361 cites W3035574324 @default.
- W3206536361 cites W3035682985 @default.
- W3206536361 cites W3035729345 @default.
- W3206536361 cites W3036304200 @default.
- W3206536361 cites W3037492894 @default.
- W3206536361 cites W3071470454 @default.
- W3206536361 cites W3089560030 @default.
- W3206536361 cites W3092171228 @default.
- W3206536361 cites W3093235747 @default.
- W3206536361 cites W3100572490 @default.
- W3206536361 cites W3103385169 @default.
- W3206536361 cites W3104224589 @default.
- W3206536361 cites W3104620392 @default.
- W3206536361 cites W3109453310 @default.
- W3206536361 cites W3120243996 @default.
- W3206536361 cites W3120460133 @default.
- W3206536361 cites W3124238039 @default.
- W3206536361 cites W3128398196 @default.
- W3206536361 cites W3162926177 @default.
- W3206536361 cites W3168115700 @default.
- W3206536361 cites W9657784 @default.
- W3206536361 hasPublicationYear "2021" @default.
- W3206536361 type Work @default.
- W3206536361 sameAs 3206536361 @default.
- W3206536361 citedByCount "0" @default.
- W3206536361 crossrefType "posted-content" @default.
- W3206536361 hasAuthorship W3206536361A5006380946 @default.
- W3206536361 hasAuthorship W3206536361A5029801689 @default.
- W3206536361 hasAuthorship W3206536361A5046296629 @default.
- W3206536361 hasAuthorship W3206536361A5051904433 @default.
- W3206536361 hasAuthorship W3206536361A5059632022 @default.