Matches in SemOpenAlex for { <https://semopenalex.org/work/W2945793108> ?p ?o ?g. }
- W2945793108 abstract "Improving the robustness of deep neural networks (DNNs) to adversarial examples is an important yet challenging problem for secure deep learning. Across existing defense techniques, adversarial training with Projected Gradient Decent (PGD) is amongst the most effective. Adversarial training solves a min-max optimization problem, with the textit{inner maximization} generating adversarial examples by maximizing the classification loss, and the textit{outer minimization} finding model parameters by minimizing the loss on adversarial examples generated from the inner maximization. A criterion that measures how well the inner maximization is solved is therefore crucial for adversarial training. In this paper, we propose such a criterion, namely First-Order Stationary Condition for constrained optimization (FOSC), to quantitatively evaluate the convergence quality of adversarial examples found in the inner maximization. With FOSC, we find that to ensure better robustness, it is essential to use adversarial examples with better convergence quality at the textit{later stages} of training. Yet at the early stages, high convergence quality adversarial examples are not necessary and may even lead to poor robustness. Based on these observations, we propose a textit{dynamic} training strategy to gradually increase the convergence quality of the generated adversarial examples, which significantly improves the robustness of adversarial training. Our theoretical and empirical results show the effectiveness of the proposed method." @default.
- W2945793108 created "2019-05-29" @default.
- W2945793108 creator A5004922473 @default.
- W2945793108 creator A5004999983 @default.
- W2945793108 creator A5030837133 @default.
- W2945793108 creator A5031013029 @default.
- W2945793108 creator A5051448391 @default.
- W2945793108 creator A5078711649 @default.
- W2945793108 date "2021-12-15" @default.
- W2945793108 modified "2023-10-01" @default.
- W2945793108 title "On the Convergence and Robustness of Adversarial Training" @default.
- W2945793108 cites W1883420340 @default.
- W2945793108 cites W2116775748 @default.
- W2945793108 cites W2118020555 @default.
- W2945793108 cites W2119112357 @default.
- W2945793108 cites W2132984949 @default.
- W2945793108 cites W2136885855 @default.
- W2945793108 cites W2180612164 @default.
- W2945793108 cites W2194775991 @default.
- W2945793108 cites W2256388387 @default.
- W2945793108 cites W2296073425 @default.
- W2945793108 cites W2603766943 @default.
- W2945793108 cites W2765384636 @default.
- W2945793108 cites W2767075075 @default.
- W2945793108 cites W2774018344 @default.
- W2945793108 cites W2787708942 @default.
- W2945793108 cites W2891229686 @default.
- W2945793108 cites W2900959181 @default.
- W2945793108 cites W2949117887 @default.
- W2945793108 cites W2963207607 @default.
- W2945793108 cites W2963249138 @default.
- W2945793108 cites W2963341956 @default.
- W2945793108 cites W2963566318 @default.
- W2945793108 cites W2963636205 @default.
- W2945793108 cites W2963744840 @default.
- W2945793108 cites W2963857521 @default.
- W2945793108 cites W2964082701 @default.
- W2945793108 cites W2964120007 @default.
- W2945793108 cites W2964153729 @default.
- W2945793108 cites W2964276371 @default.
- W2945793108 cites W2964294232 @default.
- W2945793108 cites W3217549878 @default.
- W2945793108 doi "https://doi.org/10.48550/arxiv.2112.08304" @default.
- W2945793108 hasPublicationYear "2021" @default.
- W2945793108 type Work @default.
- W2945793108 sameAs 2945793108 @default.
- W2945793108 citedByCount "101" @default.
- W2945793108 countsByYear W29457931082018 @default.
- W2945793108 countsByYear W29457931082019 @default.
- W2945793108 countsByYear W29457931082020 @default.
- W2945793108 countsByYear W29457931082021 @default.
- W2945793108 countsByYear W29457931082022 @default.
- W2945793108 crossrefType "posted-content" @default.
- W2945793108 hasAuthorship W2945793108A5004922473 @default.
- W2945793108 hasAuthorship W2945793108A5004999983 @default.
- W2945793108 hasAuthorship W2945793108A5030837133 @default.
- W2945793108 hasAuthorship W2945793108A5031013029 @default.
- W2945793108 hasAuthorship W2945793108A5051448391 @default.
- W2945793108 hasAuthorship W2945793108A5078711649 @default.
- W2945793108 hasBestOaLocation W29457931081 @default.
- W2945793108 hasConcept C104317684 @default.
- W2945793108 hasConcept C119857082 @default.
- W2945793108 hasConcept C126255220 @default.
- W2945793108 hasConcept C147764199 @default.
- W2945793108 hasConcept C154945302 @default.
- W2945793108 hasConcept C162324750 @default.
- W2945793108 hasConcept C185592680 @default.
- W2945793108 hasConcept C2776330181 @default.
- W2945793108 hasConcept C2777303404 @default.
- W2945793108 hasConcept C2984842247 @default.
- W2945793108 hasConcept C33923547 @default.
- W2945793108 hasConcept C37736160 @default.
- W2945793108 hasConcept C41008148 @default.
- W2945793108 hasConcept C50522688 @default.
- W2945793108 hasConcept C50644808 @default.
- W2945793108 hasConcept C55493867 @default.
- W2945793108 hasConcept C63479239 @default.
- W2945793108 hasConceptScore W2945793108C104317684 @default.
- W2945793108 hasConceptScore W2945793108C119857082 @default.
- W2945793108 hasConceptScore W2945793108C126255220 @default.
- W2945793108 hasConceptScore W2945793108C147764199 @default.
- W2945793108 hasConceptScore W2945793108C154945302 @default.
- W2945793108 hasConceptScore W2945793108C162324750 @default.
- W2945793108 hasConceptScore W2945793108C185592680 @default.
- W2945793108 hasConceptScore W2945793108C2776330181 @default.
- W2945793108 hasConceptScore W2945793108C2777303404 @default.
- W2945793108 hasConceptScore W2945793108C2984842247 @default.
- W2945793108 hasConceptScore W2945793108C33923547 @default.
- W2945793108 hasConceptScore W2945793108C37736160 @default.
- W2945793108 hasConceptScore W2945793108C41008148 @default.
- W2945793108 hasConceptScore W2945793108C50522688 @default.
- W2945793108 hasConceptScore W2945793108C50644808 @default.
- W2945793108 hasConceptScore W2945793108C55493867 @default.
- W2945793108 hasConceptScore W2945793108C63479239 @default.
- W2945793108 hasLocation W29457931081 @default.
- W2945793108 hasOpenAccess W2945793108 @default.
- W2945793108 hasPrimaryLocation W29457931081 @default.
- W2945793108 hasRelatedWork W2945793108 @default.
- W2945793108 hasRelatedWork W2952919291 @default.
- W2945793108 hasRelatedWork W2963857521 @default.