Matches in SemOpenAlex for { <https://semopenalex.org/work/W3176131920> ?p ?o ?g. }
- W3176131920 abstract "Deep equilibrium networks (DEQs) are a new class of models that eschews traditional depth in favor of finding the fixed point of a single nonlinear layer. These models have been shown to achieve performance competitive with the state-of-the-art deep networks while using significantly less memory. Yet they are also slower, brittle to architectural choices, and introduce potential instability to the model. In this paper, we propose a regularization scheme for DEQ models that explicitly regularizes the Jacobian of the fixed-point update equations to stabilize the learning of equilibrium models. We show that this regularization adds only minimal computational cost, significantly stabilizes the fixed-point convergence in both forward and backward passes, and scales well to high-dimensional, realistic domains (e.g., WikiText-103 language modeling and ImageNet classification). Using this method, we demonstrate, for the first time, an implicit-depth model that runs with approximately the same speed and level of performance as popular conventional deep networks such as ResNet-101, while still maintaining the constant memory footprint and architectural simplicity of DEQs. Code is available at https://github.com/locuslab/deq ." @default.
- W3176131920 created "2021-07-05" @default.
- W3176131920 creator A5006181255 @default.
- W3176131920 creator A5037832036 @default.
- W3176131920 creator A5075035644 @default.
- W3176131920 date "2021-06-27" @default.
- W3176131920 modified "2023-10-16" @default.
- W3176131920 title "Stabilizing Equilibrium Models by Jacobian Regularization" @default.
- W3176131920 cites W1542343170 @default.
- W3176131920 cites W1632114991 @default.
- W3176131920 cites W1836465849 @default.
- W3176131920 cites W1977822596 @default.
- W3176131920 cites W1990457366 @default.
- W3176131920 cites W2038281434 @default.
- W3176131920 cites W2041256929 @default.
- W3176131920 cites W2095036901 @default.
- W3176131920 cites W2108598243 @default.
- W3176131920 cites W2112594540 @default.
- W3176131920 cites W2163605009 @default.
- W3176131920 cites W2194775991 @default.
- W3176131920 cites W2340897893 @default.
- W3176131920 cites W2529714286 @default.
- W3176131920 cites W2764280570 @default.
- W3176131920 cites W2786622092 @default.
- W3176131920 cites W2795783309 @default.
- W3176131920 cites W2962832505 @default.
- W3176131920 cites W2963263347 @default.
- W3176131920 cites W2963266340 @default.
- W3176131920 cites W2963403868 @default.
- W3176131920 cites W2963446712 @default.
- W3176131920 cites W2963494889 @default.
- W3176131920 cites W2963631907 @default.
- W3176131920 cites W2963641970 @default.
- W3176131920 cites W2963685250 @default.
- W3176131920 cites W2963755523 @default.
- W3176131920 cites W2963836885 @default.
- W3176131920 cites W2963870701 @default.
- W3176131920 cites W2963970238 @default.
- W3176131920 cites W2964088127 @default.
- W3176131920 cites W2964110616 @default.
- W3176131920 cites W2964582580 @default.
- W3176131920 cites W2968953580 @default.
- W3176131920 cites W2970181183 @default.
- W3176131920 cites W2970900903 @default.
- W3176131920 cites W2973727699 @default.
- W3176131920 cites W2997347790 @default.
- W3176131920 cites W3005235017 @default.
- W3176131920 cites W3016635207 @default.
- W3176131920 cites W3027429260 @default.
- W3176131920 cites W3030163527 @default.
- W3176131920 cites W3043741336 @default.
- W3176131920 cites W3090968963 @default.
- W3176131920 cites W3094502228 @default.
- W3176131920 cites W3098967720 @default.
- W3176131920 cites W3102042883 @default.
- W3176131920 cites W3102843890 @default.
- W3176131920 cites W3104105596 @default.
- W3176131920 cites W3105113144 @default.
- W3176131920 cites W3118608800 @default.
- W3176131920 cites W3126161079 @default.
- W3176131920 cites W3127615158 @default.
- W3176131920 cites W3131747507 @default.
- W3176131920 cites W3141126219 @default.
- W3176131920 cites W3168540635 @default.
- W3176131920 doi "https://doi.org/10.48550/arxiv.2106.14342" @default.
- W3176131920 hasPublicationYear "2021" @default.
- W3176131920 type Work @default.
- W3176131920 sameAs 3176131920 @default.
- W3176131920 citedByCount "0" @default.
- W3176131920 crossrefType "posted-content" @default.
- W3176131920 hasAuthorship W3176131920A5006181255 @default.
- W3176131920 hasAuthorship W3176131920A5037832036 @default.
- W3176131920 hasAuthorship W3176131920A5075035644 @default.
- W3176131920 hasBestOaLocation W31761319201 @default.
- W3176131920 hasConcept C111919701 @default.
- W3176131920 hasConcept C11413529 @default.
- W3176131920 hasConcept C121332964 @default.
- W3176131920 hasConcept C126255220 @default.
- W3176131920 hasConcept C134306372 @default.
- W3176131920 hasConcept C154945302 @default.
- W3176131920 hasConcept C158622935 @default.
- W3176131920 hasConcept C162324750 @default.
- W3176131920 hasConcept C200331156 @default.
- W3176131920 hasConcept C2776135515 @default.
- W3176131920 hasConcept C2777303404 @default.
- W3176131920 hasConcept C28826006 @default.
- W3176131920 hasConcept C33923547 @default.
- W3176131920 hasConcept C41008148 @default.
- W3176131920 hasConcept C50522688 @default.
- W3176131920 hasConcept C61445026 @default.
- W3176131920 hasConcept C62520636 @default.
- W3176131920 hasConcept C74912251 @default.
- W3176131920 hasConcept C78045399 @default.
- W3176131920 hasConcept C94766913 @default.
- W3176131920 hasConceptScore W3176131920C111919701 @default.
- W3176131920 hasConceptScore W3176131920C11413529 @default.
- W3176131920 hasConceptScore W3176131920C121332964 @default.
- W3176131920 hasConceptScore W3176131920C126255220 @default.
- W3176131920 hasConceptScore W3176131920C134306372 @default.
- W3176131920 hasConceptScore W3176131920C154945302 @default.