FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2899692219> ?p ?o ?g. }

• W2899692219 abstract "Making neural networks robust against adversarial inputs has resulted in an arms race between new defenses and attacks. The most promising defenses, adversarially robust training and verifiably robust training, have limitations that restrict their practical applications. The adversarially robust training only makes the networks robust against a subclass of attackers and we reveal such weaknesses by developing a new attack based on interval gradients. By contrast, verifiably robust training provides protection against any L-p norm-bounded attacker but incurs orders of magnitude more computational and memory overhead than adversarially robust training. We propose two novel techniques, stochastic robust approximation and dynamic mixed training, to drastically improve the efficiency of verifiably robust training without sacrificing verified robustness. We leverage two critical insights: (1) instead of over the entire training set, sound over-approximations over randomly subsampled training data points are sufficient for efficiently guiding the robust training process; and (2) We observe that the test accuracy and verifiable robustness often conflict after certain training epochs. Therefore, we use a dynamic loss function to adaptively balance them for each epoch. We designed and implemented our techniques as part of MixTrain and evaluated it on six networks trained on three popular datasets including MNIST, CIFAR, and ImageNet-200. Our evaluations show that MixTrain can achieve up to $95.2%$ verified robust accuracy against $L_infty$ norm-bounded attackers while taking $15$ and $3$ times less training time than state-of-the-art verifiably robust training and adversarially robust training schemes, respectively. Furthermore, MixTrain easily scales to larger networks like the one trained on ImageNet-200, significantly outperforming the existing verifiably robust training methods." @default.
• W2899692219 created "2018-11-16" @default.
• W2899692219 creator A5005317598 @default.
• W2899692219 creator A5007570908 @default.
• W2899692219 creator A5016425387 @default.
• W2899692219 creator A5042401810 @default.
• W2899692219 date "2018-11-06" @default.
• W2899692219 modified "2023-09-27" @default.
• W2899692219 title "MixTrain: Scalable Training of Formally Robust Neural Networks." @default.
• W2899692219 cites W1522301498 @default.
• W2899692219 cites W1542886316 @default.
• W2899692219 cites W1673923490 @default.
• W2899692219 cites W1883420340 @default.
• W2899692219 cites W2009797711 @default.
• W2899692219 cites W2108598243 @default.
• W2899692219 cites W2112796928 @default.
• W2899692219 cites W2155195660 @default.
• W2899692219 cites W2174868984 @default.
• W2899692219 cites W2230740169 @default.
• W2899692219 cites W2243397390 @default.
• W2899692219 cites W2401231614 @default.
• W2899692219 cites W2565186948 @default.
• W2899692219 cites W2594877703 @default.
• W2899692219 cites W2603766943 @default.
• W2899692219 cites W2610190180 @default.
• W2899692219 cites W2614635052 @default.
• W2899692219 cites W2616028256 @default.
• W2899692219 cites W2619479788 @default.
• W2899692219 cites W2625220439 @default.
• W2899692219 cites W2721006554 @default.
• W2899692219 cites W2735607295 @default.
• W2899692219 cites W2765233338 @default.
• W2899692219 cites W2765384636 @default.
• W2899692219 cites W2766462876 @default.
• W2899692219 cites W2767075075 @default.
• W2899692219 cites W2767962654 @default.
• W2899692219 cites W2768899812 @default.
• W2899692219 cites W2768915615 @default.
• W2899692219 cites W2777012514 @default.
• W2899692219 cites W2783692467 @default.
• W2899692219 cites W2786118190 @default.
• W2899692219 cites W2786977288 @default.
• W2899692219 cites W2787708942 @default.
• W2899692219 cites W2788686132 @default.
• W2899692219 cites W2789524546 @default.
• W2899692219 cites W2793165286 @default.
• W2899692219 cites W2794609696 @default.
• W2899692219 cites W2799107510 @default.
• W2899692219 cites W2803392236 @default.
• W2899692219 cites W2803850896 @default.
• W2899692219 cites W2810611310 @default.
• W2899692219 cites W2889210204 @default.
• W2899692219 cites W2890660842 @default.
• W2899692219 cites W2951133631 @default.
• W2899692219 cites W2962943487 @default.
• W2899692219 cites W2963207607 @default.
• W2899692219 cites W2963327228 @default.
• W2899692219 cites W2963389226 @default.
• W2899692219 cites W2963557656 @default.
• W2899692219 cites W2963626858 @default.
• W2899692219 cites W2963744840 @default.
• W2899692219 cites W2963857521 @default.
• W2899692219 cites W2963952467 @default.
• W2899692219 cites W2964253222 @default.
• W2899692219 cites W3118608800 @default.
• W2899692219 hasPublicationYear "2018" @default.
• W2899692219 type Work @default.
• W2899692219 sameAs 2899692219 @default.
• W2899692219 citedByCount "62" @default.
• W2899692219 countsByYear W28996922192018 @default.
• W2899692219 countsByYear W28996922192019 @default.
• W2899692219 countsByYear W28996922192020 @default.
• W2899692219 countsByYear W28996922192021 @default.
• W2899692219 crossrefType "posted-content" @default.
• W2899692219 hasAuthorship W2899692219A5005317598 @default.
• W2899692219 hasAuthorship W2899692219A5007570908 @default.
• W2899692219 hasAuthorship W2899692219A5016425387 @default.
• W2899692219 hasAuthorship W2899692219A5042401810 @default.
• W2899692219 hasConcept C104317684 @default.
• W2899692219 hasConcept C119857082 @default.
• W2899692219 hasConcept C126255220 @default.
• W2899692219 hasConcept C134306372 @default.
• W2899692219 hasConcept C153083717 @default.
• W2899692219 hasConcept C154945302 @default.
• W2899692219 hasConcept C185592680 @default.
• W2899692219 hasConcept C190502265 @default.
• W2899692219 hasConcept C33923547 @default.
• W2899692219 hasConcept C34388435 @default.
• W2899692219 hasConcept C41008148 @default.
• W2899692219 hasConcept C48044578 @default.
• W2899692219 hasConcept C50644808 @default.
• W2899692219 hasConcept C55493867 @default.
• W2899692219 hasConcept C63479239 @default.
• W2899692219 hasConcept C77088390 @default.
• W2899692219 hasConceptScore W2899692219C104317684 @default.
• W2899692219 hasConceptScore W2899692219C119857082 @default.
• W2899692219 hasConceptScore W2899692219C126255220 @default.
• W2899692219 hasConceptScore W2899692219C134306372 @default.
• W2899692219 hasConceptScore W2899692219C153083717 @default.
• W2899692219 hasConceptScore W2899692219C154945302 @default.

• next