Matches in SemOpenAlex for { <https://semopenalex.org/work/W2922371350> ?p ?o ?g. }
- W2922371350 abstract "Understanding the power of depth in feed-forward neural networks is an ongoing challenge in the field of deep learning theory. While current works account for the importance of depth for the expressive power of neural-networks, it remains an open question whether these benefits are exploited during a gradient-based optimization process. In this work we explore the relation between expressivity properties of deep networks and the ability to train them efficiently using gradient-based algorithms. We give a depth separation argument for distributions with fractal structure, showing that they can be expressed efficiently by deep networks, but not with shallow ones. These distributions have a natural coarse-to-fine structure, and we show that the balance between the coarse and fine details has a crucial effect on whether the optimization process is likely to succeed. We prove that when the distribution is concentrated on the fine details, gradient-based algorithms are likely to fail. Using this result we prove that, at least in some distributions, the success of learning deep networks depends on whether the distribution can be well approximated by shallower networks, and we conjecture that this property holds in general." @default.
- W2922371350 created "2019-03-22" @default.
- W2922371350 creator A5069983576 @default.
- W2922371350 creator A5078263052 @default.
- W2922371350 date "2019-03-08" @default.
- W2922371350 modified "2023-09-27" @default.
- W2922371350 title "Is Deeper Better only when Shallow is Good" @default.
- W2922371350 cites W1522301498 @default.
- W2922371350 cites W1981530182 @default.
- W2922371350 cites W2141473882 @default.
- W2922371350 cites W2161388792 @default.
- W2922371350 cites W2208555118 @default.
- W2922371350 cites W2281746805 @default.
- W2922371350 cites W2433379750 @default.
- W2922371350 cites W2513671774 @default.
- W2922371350 cites W2550848904 @default.
- W2922371350 cites W2623127191 @default.
- W2922371350 cites W2767609493 @default.
- W2922371350 cites W2949336664 @default.
- W2922371350 cites W2952600837 @default.
- W2922371350 cites W2962742960 @default.
- W2922371350 cites W2962845550 @default.
- W2922371350 cites W2963100491 @default.
- W2922371350 cites W2964088238 @default.
- W2922371350 cites W607505555 @default.
- W2922371350 hasPublicationYear "2019" @default.
- W2922371350 type Work @default.
- W2922371350 sameAs 2922371350 @default.
- W2922371350 citedByCount "8" @default.
- W2922371350 countsByYear W29223713502019 @default.
- W2922371350 countsByYear W29223713502020 @default.
- W2922371350 countsByYear W29223713502021 @default.
- W2922371350 crossrefType "posted-content" @default.
- W2922371350 hasAuthorship W2922371350A5069983576 @default.
- W2922371350 hasAuthorship W2922371350A5078263052 @default.
- W2922371350 hasConcept C108583219 @default.
- W2922371350 hasConcept C110121322 @default.
- W2922371350 hasConcept C111472728 @default.
- W2922371350 hasConcept C111919701 @default.
- W2922371350 hasConcept C11413529 @default.
- W2922371350 hasConcept C121332964 @default.
- W2922371350 hasConcept C134306372 @default.
- W2922371350 hasConcept C138885662 @default.
- W2922371350 hasConcept C154945302 @default.
- W2922371350 hasConcept C163258240 @default.
- W2922371350 hasConcept C185592680 @default.
- W2922371350 hasConcept C189950617 @default.
- W2922371350 hasConcept C202444582 @default.
- W2922371350 hasConcept C2780990831 @default.
- W2922371350 hasConcept C2984842247 @default.
- W2922371350 hasConcept C33923547 @default.
- W2922371350 hasConcept C40636538 @default.
- W2922371350 hasConcept C41008148 @default.
- W2922371350 hasConcept C50644808 @default.
- W2922371350 hasConcept C55493867 @default.
- W2922371350 hasConcept C62520636 @default.
- W2922371350 hasConcept C9652623 @default.
- W2922371350 hasConcept C98045186 @default.
- W2922371350 hasConcept C98184364 @default.
- W2922371350 hasConceptScore W2922371350C108583219 @default.
- W2922371350 hasConceptScore W2922371350C110121322 @default.
- W2922371350 hasConceptScore W2922371350C111472728 @default.
- W2922371350 hasConceptScore W2922371350C111919701 @default.
- W2922371350 hasConceptScore W2922371350C11413529 @default.
- W2922371350 hasConceptScore W2922371350C121332964 @default.
- W2922371350 hasConceptScore W2922371350C134306372 @default.
- W2922371350 hasConceptScore W2922371350C138885662 @default.
- W2922371350 hasConceptScore W2922371350C154945302 @default.
- W2922371350 hasConceptScore W2922371350C163258240 @default.
- W2922371350 hasConceptScore W2922371350C185592680 @default.
- W2922371350 hasConceptScore W2922371350C189950617 @default.
- W2922371350 hasConceptScore W2922371350C202444582 @default.
- W2922371350 hasConceptScore W2922371350C2780990831 @default.
- W2922371350 hasConceptScore W2922371350C2984842247 @default.
- W2922371350 hasConceptScore W2922371350C33923547 @default.
- W2922371350 hasConceptScore W2922371350C40636538 @default.
- W2922371350 hasConceptScore W2922371350C41008148 @default.
- W2922371350 hasConceptScore W2922371350C50644808 @default.
- W2922371350 hasConceptScore W2922371350C55493867 @default.
- W2922371350 hasConceptScore W2922371350C62520636 @default.
- W2922371350 hasConceptScore W2922371350C9652623 @default.
- W2922371350 hasConceptScore W2922371350C98045186 @default.
- W2922371350 hasConceptScore W2922371350C98184364 @default.
- W2922371350 hasLocation W29223713501 @default.
- W2922371350 hasOpenAccess W2922371350 @default.
- W2922371350 hasPrimaryLocation W29223713501 @default.
- W2922371350 hasRelatedWork W2103496339 @default.
- W2922371350 hasRelatedWork W2161388792 @default.
- W2922371350 hasRelatedWork W2208555118 @default.
- W2922371350 hasRelatedWork W2549189808 @default.
- W2922371350 hasRelatedWork W2566079294 @default.
- W2922371350 hasRelatedWork W2605372163 @default.
- W2922371350 hasRelatedWork W2731468224 @default.
- W2922371350 hasRelatedWork W2787999646 @default.
- W2922371350 hasRelatedWork W2884368321 @default.
- W2922371350 hasRelatedWork W2926417365 @default.
- W2922371350 hasRelatedWork W2952147788 @default.
- W2922371350 hasRelatedWork W2962742960 @default.
- W2922371350 hasRelatedWork W2963982496 @default.
- W2922371350 hasRelatedWork W2964036823 @default.