Matches in SemOpenAlex for { <https://semopenalex.org/work/W3113189347> ?p ?o ?g. }
- W3113189347 endingPage "9712" @default.
- W3113189347 startingPage "9704" @default.
- W3113189347 abstract "Integrating model-based machine learning methods into deep neural architectures allows one to leverage both the expressive power of deep neural nets and the ability of model-based methods to incorporate domain-specific knowledge. In particular, many works have employed the expectation maximization (EM) algorithm in the form of an unrolled layer-wise structure that is jointly trained with a backbone neural network. However, it is difficult to discriminatively train the backbone network by backpropagating through the EM iterations as they are prone to the vanishing gradient problem. To address this issue, we propose Highway Expectation Maximization Networks (HEMNet), which is comprised of unrolled iterations of the generalized EM (GEM) algorithm based on the Newton-Rahpson method. HEMNet features scaled skip connections, or highways, along the depths of the unrolled architecture, resulting in improved gradient flow during backpropagation while incurring negligible additional computation and memory costs compared to standard unrolled EM. Furthermore, HEMNet preserves the underlying EM procedure, thereby fully retaining the convergence properties of the original EM algorithm. We achieve significant improvement in performance on several semantic segmentation benchmarks and empirically show that HEMNet effectively alleviates gradient decay." @default.
- W3113189347 created "2020-12-21" @default.
- W3113189347 creator A5027450557 @default.
- W3113189347 creator A5046475005 @default.
- W3113189347 creator A5066513515 @default.
- W3113189347 date "2021-05-18" @default.
- W3113189347 modified "2023-09-23" @default.
- W3113189347 title "Improving Gradient Flow with Unrolled Highway Expectation Maximization" @default.
- W3113189347 cites W1026270304 @default.
- W3113189347 cites W1831449718 @default.
- W3113189347 cites W1836465849 @default.
- W3113189347 cites W194242946 @default.
- W3113189347 cites W2031489346 @default.
- W3113189347 cites W2038885294 @default.
- W3113189347 cites W2049633694 @default.
- W3113189347 cites W2053742104 @default.
- W3113189347 cites W2064675550 @default.
- W3113189347 cites W2107878631 @default.
- W3113189347 cites W2116261113 @default.
- W3113189347 cites W2117539524 @default.
- W3113189347 cites W2125215748 @default.
- W3113189347 cites W2194775991 @default.
- W3113189347 cites W2302255633 @default.
- W3113189347 cites W2560023338 @default.
- W3113189347 cites W2561196672 @default.
- W3113189347 cites W2563705555 @default.
- W3113189347 cites W2567948266 @default.
- W3113189347 cites W2630837129 @default.
- W3113189347 cites W2777737607 @default.
- W3113189347 cites W2785994986 @default.
- W3113189347 cites W2798791840 @default.
- W3113189347 cites W2799166040 @default.
- W3113189347 cites W2883606943 @default.
- W3113189347 cites W2888520903 @default.
- W3113189347 cites W2892220259 @default.
- W3113189347 cites W2895340641 @default.
- W3113189347 cites W2895420332 @default.
- W3113189347 cites W2950635152 @default.
- W3113189347 cites W2955058313 @default.
- W3113189347 cites W2963403868 @default.
- W3113189347 cites W2963446712 @default.
- W3113189347 cites W2963532543 @default.
- W3113189347 cites W2963542740 @default.
- W3113189347 cites W2963727650 @default.
- W3113189347 cites W2963728677 @default.
- W3113189347 cites W2963815618 @default.
- W3113189347 cites W2964213104 @default.
- W3113189347 cites W2964309882 @default.
- W3113189347 cites W2970290486 @default.
- W3113189347 cites W2993235622 @default.
- W3113189347 cites W3035220232 @default.
- W3113189347 doi "https://doi.org/10.1609/aaai.v35i11.17167" @default.
- W3113189347 hasPublicationYear "2021" @default.
- W3113189347 type Work @default.
- W3113189347 sameAs 3113189347 @default.
- W3113189347 citedByCount "0" @default.
- W3113189347 crossrefType "journal-article" @default.
- W3113189347 hasAuthorship W3113189347A5027450557 @default.
- W3113189347 hasAuthorship W3113189347A5046475005 @default.
- W3113189347 hasAuthorship W3113189347A5066513515 @default.
- W3113189347 hasBestOaLocation W31131893471 @default.
- W3113189347 hasConcept C105795698 @default.
- W3113189347 hasConcept C11413529 @default.
- W3113189347 hasConcept C126255220 @default.
- W3113189347 hasConcept C134306372 @default.
- W3113189347 hasConcept C153083717 @default.
- W3113189347 hasConcept C154945302 @default.
- W3113189347 hasConcept C155032097 @default.
- W3113189347 hasConcept C162324750 @default.
- W3113189347 hasConcept C167879884 @default.
- W3113189347 hasConcept C182081679 @default.
- W3113189347 hasConcept C2776330181 @default.
- W3113189347 hasConcept C2777303404 @default.
- W3113189347 hasConcept C33923547 @default.
- W3113189347 hasConcept C41008148 @default.
- W3113189347 hasConcept C45374587 @default.
- W3113189347 hasConcept C49781872 @default.
- W3113189347 hasConcept C50522688 @default.
- W3113189347 hasConcept C50644808 @default.
- W3113189347 hasConceptScore W3113189347C105795698 @default.
- W3113189347 hasConceptScore W3113189347C11413529 @default.
- W3113189347 hasConceptScore W3113189347C126255220 @default.
- W3113189347 hasConceptScore W3113189347C134306372 @default.
- W3113189347 hasConceptScore W3113189347C153083717 @default.
- W3113189347 hasConceptScore W3113189347C154945302 @default.
- W3113189347 hasConceptScore W3113189347C155032097 @default.
- W3113189347 hasConceptScore W3113189347C162324750 @default.
- W3113189347 hasConceptScore W3113189347C167879884 @default.
- W3113189347 hasConceptScore W3113189347C182081679 @default.
- W3113189347 hasConceptScore W3113189347C2776330181 @default.
- W3113189347 hasConceptScore W3113189347C2777303404 @default.
- W3113189347 hasConceptScore W3113189347C33923547 @default.
- W3113189347 hasConceptScore W3113189347C41008148 @default.
- W3113189347 hasConceptScore W3113189347C45374587 @default.
- W3113189347 hasConceptScore W3113189347C49781872 @default.
- W3113189347 hasConceptScore W3113189347C50522688 @default.
- W3113189347 hasConceptScore W3113189347C50644808 @default.
- W3113189347 hasIssue "11" @default.