FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3033015882> ?p ?o ?g. }

• W3033015882 abstract "In neural networks with binary activations and or binary weights the training by gradient descent is complicated as the model has piecewise constant response. We consider stochastic binary networks, obtained by adding noises in front of activations. The expected model response becomes a smooth function of parameters, its gradient is well defined but it is challenging to estimate it accurately. We propose a new method for this estimation problem combining sampling and analytic approximation steps. The method has a significantly reduced variance at the price of a small bias which gives a very practical tradeoff in comparison with existing unbiased and biased estimators. We further show that one extra linearization step leads to a deep straight-through estimator previously known only as an ad-hoc heuristic. We experimentally show higher accuracy in gradient estimation and demonstrate a more stable and better performing training in deep convolutional models with both proposed methods." @default.
• W3033015882 created "2020-06-12" @default.
• W3033015882 creator A5068080640 @default.
• W3033015882 creator A5076376532 @default.
• W3033015882 creator A5077472871 @default.
• W3033015882 date "2020-06-04" @default.
• W3033015882 modified "2023-09-27" @default.
• W3033015882 title "Path Sample-Analytic Gradient Estimators for Stochastic Binary Networks" @default.
• W3033015882 cites W1934021597 @default.
• W3033015882 cites W2083380015 @default.
• W3033015882 cites W2095705004 @default.
• W3033015882 cites W2119717200 @default.
• W3033015882 cites W2122262818 @default.
• W3033015882 cites W2135354436 @default.
• W3033015882 cites W2187669537 @default.
• W3033015882 cites W2242818861 @default.
• W3033015882 cites W2267635276 @default.
• W3033015882 cites W2300242332 @default.
• W3033015882 cites W2319920447 @default.
• W3033015882 cites W2402098947 @default.
• W3033015882 cites W2469490737 @default.
• W3033015882 cites W2547875792 @default.
• W3033015882 cites W2602076750 @default.
• W3033015882 cites W2604700561 @default.
• W3033015882 cites W2629785709 @default.
• W3033015882 cites W2740840795 @default.
• W3033015882 cites W2783525259 @default.
• W3033015882 cites W2795358072 @default.
• W3033015882 cites W2890984855 @default.
• W3033015882 cites W2903462692 @default.
• W3033015882 cites W2907087576 @default.
• W3033015882 cites W2909637611 @default.
• W3033015882 cites W2920112539 @default.
• W3033015882 cites W2952165242 @default.
• W3033015882 cites W2963619462 @default.
• W3033015882 cites W2963671426 @default.
• W3033015882 cites W2963675378 @default.
• W3033015882 cites W2963851840 @default.
• W3033015882 cites W2963891249 @default.
• W3033015882 cites W2964339591 @default.
• W3033015882 cites W2970971581 @default.
• W3033015882 cites W2977484217 @default.
• W3033015882 cites W3022571137 @default.
• W3033015882 hasPublicationYear "2020" @default.
• W3033015882 type Work @default.
• W3033015882 sameAs 3033015882 @default.
• W3033015882 citedByCount "0" @default.
• W3033015882 crossrefType "posted-content" @default.
• W3033015882 hasAuthorship W3033015882A5068080640 @default.
• W3033015882 hasAuthorship W3033015882A5076376532 @default.
• W3033015882 hasAuthorship W3033015882A5077472871 @default.
• W3033015882 hasConcept C105795698 @default.
• W3033015882 hasConcept C11210021 @default.
• W3033015882 hasConcept C11413529 @default.
• W3033015882 hasConcept C121332964 @default.
• W3033015882 hasConcept C121955636 @default.
• W3033015882 hasConcept C126255220 @default.
• W3033015882 hasConcept C134306372 @default.
• W3033015882 hasConcept C144133560 @default.
• W3033015882 hasConcept C153258448 @default.
• W3033015882 hasConcept C154945302 @default.
• W3033015882 hasConcept C158622935 @default.
• W3033015882 hasConcept C164660894 @default.
• W3033015882 hasConcept C185429906 @default.
• W3033015882 hasConcept C19499675 @default.
• W3033015882 hasConcept C196083921 @default.
• W3033015882 hasConcept C199360897 @default.
• W3033015882 hasConcept C206688291 @default.
• W3033015882 hasConcept C2777027219 @default.
• W3033015882 hasConcept C28826006 @default.
• W3033015882 hasConcept C33923547 @default.
• W3033015882 hasConcept C41008148 @default.
• W3033015882 hasConcept C48372109 @default.
• W3033015882 hasConcept C50644808 @default.
• W3033015882 hasConcept C52740198 @default.
• W3033015882 hasConcept C62520636 @default.
• W3033015882 hasConcept C94375191 @default.
• W3033015882 hasConceptScore W3033015882C105795698 @default.
• W3033015882 hasConceptScore W3033015882C11210021 @default.
• W3033015882 hasConceptScore W3033015882C11413529 @default.
• W3033015882 hasConceptScore W3033015882C121332964 @default.
• W3033015882 hasConceptScore W3033015882C121955636 @default.
• W3033015882 hasConceptScore W3033015882C126255220 @default.
• W3033015882 hasConceptScore W3033015882C134306372 @default.
• W3033015882 hasConceptScore W3033015882C144133560 @default.
• W3033015882 hasConceptScore W3033015882C153258448 @default.
• W3033015882 hasConceptScore W3033015882C154945302 @default.
• W3033015882 hasConceptScore W3033015882C158622935 @default.
• W3033015882 hasConceptScore W3033015882C164660894 @default.
• W3033015882 hasConceptScore W3033015882C185429906 @default.
• W3033015882 hasConceptScore W3033015882C19499675 @default.
• W3033015882 hasConceptScore W3033015882C196083921 @default.
• W3033015882 hasConceptScore W3033015882C199360897 @default.
• W3033015882 hasConceptScore W3033015882C206688291 @default.
• W3033015882 hasConceptScore W3033015882C2777027219 @default.
• W3033015882 hasConceptScore W3033015882C28826006 @default.
• W3033015882 hasConceptScore W3033015882C33923547 @default.
• W3033015882 hasConceptScore W3033015882C41008148 @default.
• W3033015882 hasConceptScore W3033015882C48372109 @default.
• W3033015882 hasConceptScore W3033015882C50644808 @default.

• next