Matches in SemOpenAlex for { <https://semopenalex.org/work/W3156066263> ?p ?o ?g. }
- W3156066263 abstract "Recent work has highlighted several advantages of enforcing orthogonality in the weight layers of deep networks, such as maintaining the stability of activations, preserving gradient norms, and enhancing adversarial robustness by enforcing low Lipschitz constants. Although numerous methods exist for enforcing the orthogonality of fully-connected layers, those for convolutional layers are more heuristic in nature, often focusing on penalty methods or limited classes of convolutions. In this work, we propose and evaluate an alternative approach to directly parameterize convolutional layers that are constrained to be orthogonal. Specifically, we propose to apply the Cayley transform to a skew-symmetric convolution in the Fourier domain, so that the inverse convolution needed by the Cayley transform can be computed efficiently. We compare our method to previous Lipschitz-constrained and orthogonal convolutional layers and show that it indeed preserves orthogonality to a high degree even for large convolutions. Applied to the problem of certified adversarial robustness, we show that networks incorporating the layer outperform existing deterministic methods for certified defense against $ell_2$-norm-bounded adversaries, while scaling to larger architectures than previously investigated. Code is available at this https URL." @default.
- W3156066263 created "2021-04-26" @default.
- W3156066263 creator A5049077447 @default.
- W3156066263 creator A5075035644 @default.
- W3156066263 date "2021-04-14" @default.
- W3156066263 modified "2023-09-27" @default.
- W3156066263 title "Orthogonalizing Convolutional Layers with the Cayley Transform" @default.
- W3156066263 cites W1522301498 @default.
- W3156066263 cites W1542938076 @default.
- W3156066263 cites W1553137900 @default.
- W3156066263 cites W1804110266 @default.
- W3156066263 cites W2054276177 @default.
- W3156066263 cites W2100115174 @default.
- W3156066263 cites W2125930537 @default.
- W3156066263 cites W2194775991 @default.
- W3156066263 cites W2401231614 @default.
- W3156066263 cites W2543296129 @default.
- W3156066263 cites W2553902701 @default.
- W3156066263 cites W2587753047 @default.
- W3156066263 cites W2721006554 @default.
- W3156066263 cites W2768915615 @default.
- W3156066263 cites W2786977288 @default.
- W3156066263 cites W2796892552 @default.
- W3156066263 cites W2899962223 @default.
- W3156066263 cites W2907121943 @default.
- W3156066263 cites W2911634294 @default.
- W3156066263 cites W2914344989 @default.
- W3156066263 cites W2962851953 @default.
- W3156066263 cites W2962943487 @default.
- W3156066263 cites W2962949867 @default.
- W3156066263 cites W2963042606 @default.
- W3156066263 cites W2963054787 @default.
- W3156066263 cites W2963103976 @default.
- W3156066263 cites W2963172539 @default.
- W3156066263 cites W2963424732 @default.
- W3156066263 cites W2963496101 @default.
- W3156066263 cites W2963540169 @default.
- W3156066263 cites W2963556353 @default.
- W3156066263 cites W2963570896 @default.
- W3156066263 cites W2963600714 @default.
- W3156066263 cites W2963626025 @default.
- W3156066263 cites W2963664410 @default.
- W3156066263 cites W2963685250 @default.
- W3156066263 cites W2963836885 @default.
- W3156066263 cites W2963857521 @default.
- W3156066263 cites W2964065616 @default.
- W3156066263 cites W2964153729 @default.
- W3156066263 cites W2964294232 @default.
- W3156066263 cites W2970615870 @default.
- W3156066263 cites W2970620827 @default.
- W3156066263 cites W2974500493 @default.
- W3156066263 cites W2995455260 @default.
- W3156066263 cites W2996609754 @default.
- W3156066263 cites W3012913016 @default.
- W3156066263 cites W3016506115 @default.
- W3156066263 cites W3034351824 @default.
- W3156066263 cites W3034994123 @default.
- W3156066263 cites W3101204503 @default.
- W3156066263 hasPublicationYear "2021" @default.
- W3156066263 type Work @default.
- W3156066263 sameAs 3156066263 @default.
- W3156066263 citedByCount "2" @default.
- W3156066263 countsByYear W31560662632021 @default.
- W3156066263 crossrefType "posted-content" @default.
- W3156066263 hasAuthorship W3156066263A5049077447 @default.
- W3156066263 hasAuthorship W3156066263A5075035644 @default.
- W3156066263 hasConcept C104317684 @default.
- W3156066263 hasConcept C11413529 @default.
- W3156066263 hasConcept C126255220 @default.
- W3156066263 hasConcept C134306372 @default.
- W3156066263 hasConcept C154945302 @default.
- W3156066263 hasConcept C17137986 @default.
- W3156066263 hasConcept C185592680 @default.
- W3156066263 hasConcept C202444582 @default.
- W3156066263 hasConcept C22324862 @default.
- W3156066263 hasConcept C2524010 @default.
- W3156066263 hasConcept C33923547 @default.
- W3156066263 hasConcept C34388435 @default.
- W3156066263 hasConcept C41008148 @default.
- W3156066263 hasConcept C45347329 @default.
- W3156066263 hasConcept C50644808 @default.
- W3156066263 hasConcept C55493867 @default.
- W3156066263 hasConcept C63479239 @default.
- W3156066263 hasConceptScore W3156066263C104317684 @default.
- W3156066263 hasConceptScore W3156066263C11413529 @default.
- W3156066263 hasConceptScore W3156066263C126255220 @default.
- W3156066263 hasConceptScore W3156066263C134306372 @default.
- W3156066263 hasConceptScore W3156066263C154945302 @default.
- W3156066263 hasConceptScore W3156066263C17137986 @default.
- W3156066263 hasConceptScore W3156066263C185592680 @default.
- W3156066263 hasConceptScore W3156066263C202444582 @default.
- W3156066263 hasConceptScore W3156066263C22324862 @default.
- W3156066263 hasConceptScore W3156066263C2524010 @default.
- W3156066263 hasConceptScore W3156066263C33923547 @default.
- W3156066263 hasConceptScore W3156066263C34388435 @default.
- W3156066263 hasConceptScore W3156066263C41008148 @default.
- W3156066263 hasConceptScore W3156066263C45347329 @default.
- W3156066263 hasConceptScore W3156066263C50644808 @default.
- W3156066263 hasConceptScore W3156066263C55493867 @default.
- W3156066263 hasConceptScore W3156066263C63479239 @default.