Matches in SemOpenAlex for { <https://semopenalex.org/work/W2969479370> ?p ?o ?g. }
- W2969479370 abstract "Training Deep Neural Networks (DNNs) that are robust to norm bounded adversarial attacks remains an elusive problem. While verification based methods are generally too expensive to robustly train large networks, it was demonstrated in Gowal et al that bounded input intervals can be inexpensively propagated per layer through large networks. This interval bound propagation (IBP) approach led to high robustness and was the first to be employed on large networks. However, due to the very loose nature of the IBP bounds, particularly for large networks, the required training procedure is complex and involved. In this paper, we closely examine the bounds of a block of layers composed of an affine layer followed by a ReLU nonlinearity followed by another affine layer. To this end, we propose expected bounds, true bounds in expectation, that are provably tighter than IBP bounds in expectation. We then extend this result to deeper networks through blockwise propagation and show that we can achieve orders of magnitudes tighter bounds compared to IBP. With such tight bounds, we demonstrate that a simple standard training procedure can achieve the best robustness-accuracy trade-off across several architectures on both MNIST and CIFAR10." @default.
- W2969479370 created "2019-08-29" @default.
- W2969479370 creator A5024763828 @default.
- W2969479370 creator A5031191315 @default.
- W2969479370 creator A5048234926 @default.
- W2969479370 creator A5069059224 @default.
- W2969479370 creator A5091670711 @default.
- W2969479370 date "2019-05-28" @default.
- W2969479370 modified "2023-09-27" @default.
- W2969479370 title "Expected Tight Bounds for Robust Deep Neural Network Training" @default.
- W2969479370 cites W1821462560 @default.
- W2969479370 cites W1903029394 @default.
- W2969479370 cites W2160815625 @default.
- W2969479370 cites W2194775991 @default.
- W2969479370 cites W2243397390 @default.
- W2969479370 cites W2543927648 @default.
- W2969479370 cites W2594877703 @default.
- W2969479370 cites W2721006554 @default.
- W2969479370 cites W2786977288 @default.
- W2969479370 cites W2799237514 @default.
- W2969479370 cites W2803850896 @default.
- W2969479370 cites W2890660842 @default.
- W2969479370 cites W2899692219 @default.
- W2969479370 cites W2899771611 @default.
- W2969479370 cites W2902560707 @default.
- W2969479370 cites W2941142343 @default.
- W2969479370 cites W2962943487 @default.
- W2969479370 cites W2963207607 @default.
- W2969479370 cites W2963496101 @default.
- W2969479370 cites W2963735478 @default.
- W2969479370 hasPublicationYear "2019" @default.
- W2969479370 type Work @default.
- W2969479370 sameAs 2969479370 @default.
- W2969479370 citedByCount "0" @default.
- W2969479370 crossrefType "posted-content" @default.
- W2969479370 hasAuthorship W2969479370A5024763828 @default.
- W2969479370 hasAuthorship W2969479370A5031191315 @default.
- W2969479370 hasAuthorship W2969479370A5048234926 @default.
- W2969479370 hasAuthorship W2969479370A5069059224 @default.
- W2969479370 hasAuthorship W2969479370A5091670711 @default.
- W2969479370 hasConcept C104317684 @default.
- W2969479370 hasConcept C11413529 @default.
- W2969479370 hasConcept C121332964 @default.
- W2969479370 hasConcept C134306372 @default.
- W2969479370 hasConcept C154945302 @default.
- W2969479370 hasConcept C155032097 @default.
- W2969479370 hasConcept C158622935 @default.
- W2969479370 hasConcept C17744445 @default.
- W2969479370 hasConcept C185592680 @default.
- W2969479370 hasConcept C190502265 @default.
- W2969479370 hasConcept C191795146 @default.
- W2969479370 hasConcept C199539241 @default.
- W2969479370 hasConcept C202444582 @default.
- W2969479370 hasConcept C2984842247 @default.
- W2969479370 hasConcept C33923547 @default.
- W2969479370 hasConcept C34388435 @default.
- W2969479370 hasConcept C41008148 @default.
- W2969479370 hasConcept C50644808 @default.
- W2969479370 hasConcept C55493867 @default.
- W2969479370 hasConcept C62520636 @default.
- W2969479370 hasConcept C63479239 @default.
- W2969479370 hasConcept C77553402 @default.
- W2969479370 hasConcept C92757383 @default.
- W2969479370 hasConceptScore W2969479370C104317684 @default.
- W2969479370 hasConceptScore W2969479370C11413529 @default.
- W2969479370 hasConceptScore W2969479370C121332964 @default.
- W2969479370 hasConceptScore W2969479370C134306372 @default.
- W2969479370 hasConceptScore W2969479370C154945302 @default.
- W2969479370 hasConceptScore W2969479370C155032097 @default.
- W2969479370 hasConceptScore W2969479370C158622935 @default.
- W2969479370 hasConceptScore W2969479370C17744445 @default.
- W2969479370 hasConceptScore W2969479370C185592680 @default.
- W2969479370 hasConceptScore W2969479370C190502265 @default.
- W2969479370 hasConceptScore W2969479370C191795146 @default.
- W2969479370 hasConceptScore W2969479370C199539241 @default.
- W2969479370 hasConceptScore W2969479370C202444582 @default.
- W2969479370 hasConceptScore W2969479370C2984842247 @default.
- W2969479370 hasConceptScore W2969479370C33923547 @default.
- W2969479370 hasConceptScore W2969479370C34388435 @default.
- W2969479370 hasConceptScore W2969479370C41008148 @default.
- W2969479370 hasConceptScore W2969479370C50644808 @default.
- W2969479370 hasConceptScore W2969479370C55493867 @default.
- W2969479370 hasConceptScore W2969479370C62520636 @default.
- W2969479370 hasConceptScore W2969479370C63479239 @default.
- W2969479370 hasConceptScore W2969479370C77553402 @default.
- W2969479370 hasConceptScore W2969479370C92757383 @default.
- W2969479370 hasLocation W29694793701 @default.
- W2969479370 hasOpenAccess W2969479370 @default.
- W2969479370 hasPrimaryLocation W29694793701 @default.
- W2969479370 hasRelatedWork W1988445563 @default.
- W2969479370 hasRelatedWork W2026496773 @default.
- W2969479370 hasRelatedWork W2054854199 @default.
- W2969479370 hasRelatedWork W2106466811 @default.
- W2969479370 hasRelatedWork W2400225133 @default.
- W2969479370 hasRelatedWork W2551251609 @default.
- W2969479370 hasRelatedWork W2889230864 @default.
- W2969479370 hasRelatedWork W2923673691 @default.
- W2969479370 hasRelatedWork W2949670066 @default.
- W2969479370 hasRelatedWork W2951316532 @default.
- W2969479370 hasRelatedWork W2953128281 @default.