Matches in SemOpenAlex for { <https://semopenalex.org/work/W3198977492> ?p ?o ?g. }
- W3198977492 abstract "Joint Energy-based Model (JEM) is a recently proposed hybrid model that retains strong discriminative power of modern CNN classifiers, while generating samples rivaling the quality of GAN-based approaches. In this paper, we propose a variety of new training procedures and architecture features to improve JEM's accuracy, training stability, and speed altogether. 1) We propose a proximal SGLD to generate samples in the proximity of samples from the previous step, which improves the stability. 2) We further treat the approximate maximum likelihood learning of EBM as a multi-step differential game, and extend the YOPO framework to cut out redundant calculations during backpropagation, which accelerates the training substantially. 3) Rather than initializing SGLD chain from random noise, we introduce a new informative initialization that samples from a distribution estimated from training data. 4) This informative initialization allows us to enable batch normalization in JEM, which further releases the power of modern CNN architectures for hybrid modeling. Code: https://github.com/sndnyang/JEMPP" @default.
- W3198977492 created "2021-09-27" @default.
- W3198977492 creator A5036338045 @default.
- W3198977492 creator A5069485589 @default.
- W3198977492 date "2021-09-18" @default.
- W3198977492 modified "2023-09-27" @default.
- W3198977492 title "JEM++: Improved Techniques for Training JEM" @default.
- W3198977492 cites W1479807131 @default.
- W3198977492 cites W1587799944 @default.
- W3198977492 cites W1836465849 @default.
- W3198977492 cites W1959608418 @default.
- W3198977492 cites W1994616650 @default.
- W3198977492 cites W2013035813 @default.
- W3198977492 cites W2049633694 @default.
- W3198977492 cites W2099471712 @default.
- W3198977492 cites W2116064496 @default.
- W3198977492 cites W2116825644 @default.
- W3198977492 cites W2135779729 @default.
- W3198977492 cites W2152790380 @default.
- W3198977492 cites W2161914416 @default.
- W3198977492 cites W2167433878 @default.
- W3198977492 cites W2194775991 @default.
- W3198977492 cites W2272361785 @default.
- W3198977492 cites W2335728318 @default.
- W3198977492 cites W2503523779 @default.
- W3198977492 cites W2531327146 @default.
- W3198977492 cites W2735607295 @default.
- W3198977492 cites W2765384636 @default.
- W3198977492 cites W2893749619 @default.
- W3198977492 cites W2922772346 @default.
- W3198977492 cites W2963139417 @default.
- W3198977492 cites W2963184668 @default.
- W3198977492 cites W2963207607 @default.
- W3198977492 cites W2963540976 @default.
- W3198977492 cites W2963546708 @default.
- W3198977492 cites W2963836885 @default.
- W3198977492 cites W2963981733 @default.
- W3198977492 cites W2964121744 @default.
- W3198977492 cites W2964137095 @default.
- W3198977492 cites W2964153729 @default.
- W3198977492 cites W2964212410 @default.
- W3198977492 cites W2964253222 @default.
- W3198977492 cites W2964335397 @default.
- W3198977492 cites W2970316625 @default.
- W3198977492 cites W2970641149 @default.
- W3198977492 cites W2971034910 @default.
- W3198977492 cites W2979473621 @default.
- W3198977492 cites W2989576588 @default.
- W3198977492 cites W2995085126 @default.
- W3198977492 cites W2996629283 @default.
- W3198977492 cites W2998462233 @default.
- W3198977492 cites W3035166812 @default.
- W3198977492 cites W3035743198 @default.
- W3198977492 cites W3094191864 @default.
- W3198977492 cites W3098510582 @default.
- W3198977492 cites W3098903812 @default.
- W3198977492 cites W3100572490 @default.
- W3198977492 cites W3105983217 @default.
- W3198977492 cites W3118608800 @default.
- W3198977492 cites W3119691539 @default.
- W3198977492 hasPublicationYear "2021" @default.
- W3198977492 type Work @default.
- W3198977492 sameAs 3198977492 @default.
- W3198977492 citedByCount "0" @default.
- W3198977492 crossrefType "posted-content" @default.
- W3198977492 hasAuthorship W3198977492A5036338045 @default.
- W3198977492 hasAuthorship W3198977492A5069485589 @default.
- W3198977492 hasBestOaLocation W31989774921 @default.
- W3198977492 hasConcept C112972136 @default.
- W3198977492 hasConcept C114466953 @default.
- W3198977492 hasConcept C119857082 @default.
- W3198977492 hasConcept C136886441 @default.
- W3198977492 hasConcept C144024400 @default.
- W3198977492 hasConcept C154945302 @default.
- W3198977492 hasConcept C155032097 @default.
- W3198977492 hasConcept C177264268 @default.
- W3198977492 hasConcept C19165224 @default.
- W3198977492 hasConcept C199360897 @default.
- W3198977492 hasConcept C2776760102 @default.
- W3198977492 hasConcept C41008148 @default.
- W3198977492 hasConcept C50644808 @default.
- W3198977492 hasConcept C97931131 @default.
- W3198977492 hasConceptScore W3198977492C112972136 @default.
- W3198977492 hasConceptScore W3198977492C114466953 @default.
- W3198977492 hasConceptScore W3198977492C119857082 @default.
- W3198977492 hasConceptScore W3198977492C136886441 @default.
- W3198977492 hasConceptScore W3198977492C144024400 @default.
- W3198977492 hasConceptScore W3198977492C154945302 @default.
- W3198977492 hasConceptScore W3198977492C155032097 @default.
- W3198977492 hasConceptScore W3198977492C177264268 @default.
- W3198977492 hasConceptScore W3198977492C19165224 @default.
- W3198977492 hasConceptScore W3198977492C199360897 @default.
- W3198977492 hasConceptScore W3198977492C2776760102 @default.
- W3198977492 hasConceptScore W3198977492C41008148 @default.
- W3198977492 hasConceptScore W3198977492C50644808 @default.
- W3198977492 hasConceptScore W3198977492C97931131 @default.
- W3198977492 hasLocation W31989774921 @default.
- W3198977492 hasOpenAccess W3198977492 @default.
- W3198977492 hasPrimaryLocation W31989774921 @default.
- W3198977492 hasRelatedWork W10944326 @default.