Matches in SemOpenAlex for { <https://semopenalex.org/work/W3129119905> ?p ?o ?g. }
Showing items 1 to 79 of
79
with 100 items per page.
- W3129119905 abstract "The vulnerability of deep networks to adversarial attacks is a central problem for deep learning from the perspective of both cognition and security. The current most successful defense method is to train a classifier using adversarial images created during learning. Another defense approach involves transformation or purification of the original input to remove adversarial signals before the image is classified. We focus on defending naturally-trained classifiers using Markov Chain Monte Carlo (MCMC) sampling with an Energy-Based Model (EBM) for adversarial purification. In contrast to adversarial training, our approach is intended to secure pre-existing and highly vulnerable classifiers. The memoryless behavior of long-run MCMC sampling will eventually remove adversarial signals, while metastable behavior preserves consistent appearance of MCMC samples after many steps to allow accurate long-run prediction. Balancing these factors can lead to effective purification and robust classification. We evaluate adversarial defense with an EBM using the strongest known attacks against purification. Our contributions are 1) an improved method for training EBM's with realistic long-run MCMC samples, 2) an Expectation-Over-Transformation (EOT) defense that resolves theoretical ambiguities for stochastic defenses and from which the EOT attack naturally follows, and 3) state-of-the-art adversarial defense for naturally-trained classifiers and competitive defense compared to adversarially-trained classifiers on Cifar-10, SVHN, and Cifar-100. Code and pre-trained models are available at https://github.com/point0bar1/ebm-defense." @default.
- W3129119905 created "2021-02-15" @default.
- W3129119905 creator A5034228010 @default.
- W3129119905 creator A5037917365 @default.
- W3129119905 creator A5048502859 @default.
- W3129119905 date "2020-05-27" @default.
- W3129119905 modified "2023-10-18" @default.
- W3129119905 title "Stochastic Security: Adversarial Defense Using Long-Run Dynamics of Energy-Based Models" @default.
- W3129119905 cites W1592998450 @default.
- W3129119905 cites W1997584208 @default.
- W3129119905 cites W2116064496 @default.
- W3129119905 cites W2150920547 @default.
- W3129119905 cites W2194775991 @default.
- W3129119905 cites W2401231614 @default.
- W3129119905 cites W2765233338 @default.
- W3129119905 cites W2786118190 @default.
- W3129119905 cites W2787496614 @default.
- W3129119905 cites W2787708942 @default.
- W3129119905 cites W2787733970 @default.
- W3129119905 cites W2922772346 @default.
- W3129119905 cites W2946316836 @default.
- W3129119905 cites W2949457404 @default.
- W3129119905 cites W2950048339 @default.
- W3129119905 cites W2954978443 @default.
- W3129119905 cites W2963207607 @default.
- W3129119905 cites W2963485691 @default.
- W3129119905 cites W2963920068 @default.
- W3129119905 cites W2964121744 @default.
- W3129119905 cites W2964253222 @default.
- W3129119905 cites W2970680991 @default.
- W3129119905 cites W2980728855 @default.
- W3129119905 cites W2995085126 @default.
- W3129119905 cites W2998462233 @default.
- W3129119905 cites W3034994123 @default.
- W3129119905 cites W3103340107 @default.
- W3129119905 doi "https://doi.org/10.48550/arxiv.2005.13525" @default.
- W3129119905 hasPublicationYear "2020" @default.
- W3129119905 type Work @default.
- W3129119905 sameAs 3129119905 @default.
- W3129119905 citedByCount "1" @default.
- W3129119905 countsByYear W31291199052021 @default.
- W3129119905 crossrefType "posted-content" @default.
- W3129119905 hasAuthorship W3129119905A5034228010 @default.
- W3129119905 hasAuthorship W3129119905A5037917365 @default.
- W3129119905 hasAuthorship W3129119905A5048502859 @default.
- W3129119905 hasBestOaLocation W31291199051 @default.
- W3129119905 hasConcept C107673813 @default.
- W3129119905 hasConcept C108583219 @default.
- W3129119905 hasConcept C111350023 @default.
- W3129119905 hasConcept C119857082 @default.
- W3129119905 hasConcept C154945302 @default.
- W3129119905 hasConcept C37736160 @default.
- W3129119905 hasConcept C41008148 @default.
- W3129119905 hasConcept C95623464 @default.
- W3129119905 hasConceptScore W3129119905C107673813 @default.
- W3129119905 hasConceptScore W3129119905C108583219 @default.
- W3129119905 hasConceptScore W3129119905C111350023 @default.
- W3129119905 hasConceptScore W3129119905C119857082 @default.
- W3129119905 hasConceptScore W3129119905C154945302 @default.
- W3129119905 hasConceptScore W3129119905C37736160 @default.
- W3129119905 hasConceptScore W3129119905C41008148 @default.
- W3129119905 hasConceptScore W3129119905C95623464 @default.
- W3129119905 hasLocation W31291199051 @default.
- W3129119905 hasOpenAccess W3129119905 @default.
- W3129119905 hasPrimaryLocation W31291199051 @default.
- W3129119905 hasRelatedWork W2922457425 @default.
- W3129119905 hasRelatedWork W2961085424 @default.
- W3129119905 hasRelatedWork W2991087447 @default.
- W3129119905 hasRelatedWork W3007495838 @default.
- W3129119905 hasRelatedWork W3009460750 @default.
- W3129119905 hasRelatedWork W3014300295 @default.
- W3129119905 hasRelatedWork W3079760979 @default.
- W3129119905 hasRelatedWork W3164822677 @default.
- W3129119905 hasRelatedWork W4223943233 @default.
- W3129119905 hasRelatedWork W4250304930 @default.
- W3129119905 isParatext "false" @default.
- W3129119905 isRetracted "false" @default.
- W3129119905 magId "3129119905" @default.
- W3129119905 workType "article" @default.