Підтримка
www.wikidata.uk-ua.nina.az
Ne plutati z Peredavalna funkciya Funkciya aktivaciyi abo peredavalna funkciya angl activation function takozh excitation function squashing function transfer function shtuchnogo nejrona zalezhnist vihidnogo signalu shtuchnogo nejrona vid vhidnogo Zazvichaj peredavalna funkciya ϕ x displaystyle phi x vidobrazhaye dijsni chisla na interval 1 1 displaystyle 1 1 abo 0 1 displaystyle 0 1 Bilshist vidiv nejronnih merezh dlya funkciyi aktivaciyi vikoristovuyut sigmoyidi ADALINE i samoorganizacijni karti vikoristovuyut linijni funkciyi aktivaciyi a radialno bazisni merezhi vikoristovuyut radialni bazisni funkciyi Matematichno dovedeno sho trisharovij perceptron z vikoristannyam sigmoyidnoyi funkciyi aktivaciyi mozhe aproksimuvati bud yaku neperervnu funkciyu z dovilnoyu tochnistyu Teorema Cibenka Metod zvorotnogo poshirennya pomilki vimagaye shob funkciya aktivaciyi bula neperervnoyu nelinijnoyu monotonno zrostayuchoyu i diferencijovnoyu V zadachi en klasifikaciyi nejroni ostannogo sharu zazvichaj vikoristovuyut softmax yak funkciyu aktivaciyi U hemometrici funkciya yaka vikoristovuyetsya v metodi nejronnoyi sitki dlya peretvorennya u vuzlah vhidnih danih z bud yakoyi oblasti znachen zokrema neperervnih u chitko okreslenij ryad znachen napr v 0 chi 1 Porivnyannya peredavalnih funkcijDeyaki bazhani vlastivosti peredavalnoyi funkciyi vklyuchayut Nelinijna koli peredavalna funkciya nelinijna to yak dovedeno dvosharova nejronna merezha ye universalnoyu aproksimaciyeyu funkcij Totozhna peredavalna funkciya ne maye takoyi vlastivosti Koli dekilka shariv vikoristovuyut totozhnu peredavalnu funkciyu todi vsya merezha ekvivalentna odnosharovij modeli Neperervna diferencijovnist cya vlastivist bazhana RELU ne ye neperervno diferencijovnoyu i maye neodnoznachne rishennya dlya optimizaciyi zasnovanij na gradiyenti dlya vikoristannya metodiv optimizaciyi zasnovanih na gradiyenti Peredavalna funkciya dvijkovij krok ne diferencijovna u 0 ale diferencijovna v usih inshih znachennya sho ye problemoyu dlya metodiv zasnovanih na gradiyenti Oblast viznachennya Monotonnist Gladka funkciya z monotonnoyu pohidnoyu Nablizhennya do totozhnoyi funkciyi f x x displaystyle f x x v pochatku koordinat U nastupnij tablici porivnyuyutsya deyaki peredavalni funkciyi vid odniyeyi zminnoyi x z poperednogo sharu Nazva Grafik Rivnyannya Pohidna po x Oblast Poryadok gladkosti Monotonnist Monotonnist pohidnoyi Nablizhennya do Totozhnoyi funkciyi v pochatku koordinat Totozhna f x x displaystyle f x x f x 1 displaystyle f x 1 displaystyle infty infty C displaystyle C infty Tak Tak Tak Dvijkovij krok f x 0 for x lt 0 1 for x 0 displaystyle f x begin cases 0 amp text for x lt 0 1 amp text for x geqslant 0 end cases f x 0 for x 0 for x 0 displaystyle f x begin cases 0 amp text for x neq 0 amp text for x 0 end cases 0 1 displaystyle 0 1 C 1 displaystyle C 1 Tak Ni Ni Logistichna a k a Sigmoyida abo M yakij krok f x s x 1 1 e x displaystyle f x sigma x frac 1 1 e x 1 f x f x 1 f x displaystyle f x f x 1 f x 0 1 displaystyle 0 1 C displaystyle C infty Tak Ni Ni TanH f x tanh x e x e x e x e x displaystyle f x tanh x frac e x e x e x e x f x 1 f x 2 displaystyle f x 1 f x 2 1 1 displaystyle 1 1 C displaystyle C infty Tak Ni Tak ArcTan f x tan 1 x displaystyle f x tan 1 x f x 1 x 2 1 displaystyle f x frac 1 x 2 1 p 2 p 2 displaystyle left frac pi 2 frac pi 2 right C displaystyle C infty Tak Ni Tak Softsign f x x 1 x displaystyle f x frac x 1 x f x 1 1 x 2 displaystyle f x frac 1 1 x 2 1 1 displaystyle 1 1 C 1 displaystyle C 1 Tak Ni Tak Inverse square root unit ISRU f x x 1 a x 2 displaystyle f x frac x sqrt 1 alpha x 2 f x 1 1 a x 2 3 displaystyle f x left frac 1 sqrt 1 alpha x 2 right 3 1 a 1 a displaystyle left frac 1 sqrt alpha frac 1 sqrt alpha right C displaystyle C infty Tak Ni Tak Vipryamlena linijna Rectified linear unit ReLU f x 0 for x lt 0 x for x 0 displaystyle f x begin cases 0 amp text for x lt 0 x amp text for x geqslant 0 end cases f x 0 for x lt 0 1 for x 0 displaystyle f x begin cases 0 amp text for x lt 0 1 amp text for x geqslant 0 end cases 0 displaystyle 0 infty C 0 displaystyle C 0 Tak Tak Ni Leaky rectified linear unit Leaky ReLU f x 0 01 x for x lt 0 x for x 0 displaystyle f x begin cases 0 01x amp text for x lt 0 x amp text for x geqslant 0 end cases f x 0 01 for x lt 0 1 for x 0 displaystyle f x begin cases 0 01 amp text for x lt 0 1 amp text for x geqslant 0 end cases displaystyle infty infty C 0 displaystyle C 0 Tak Tak Ni Parameteric rectified linear unit PReLU f a x a x for x lt 0 x for x 0 displaystyle f alpha x begin cases alpha x amp text for x lt 0 x amp text for x geqslant 0 end cases f a x a for x lt 0 1 for x 0 displaystyle f alpha x begin cases alpha amp text for x lt 0 1 amp text for x geqslant 0 end cases displaystyle infty infty 2 C 0 displaystyle C 0 Tak a 0 displaystyle alpha geqslant 0 Tak Tak a 1 displaystyle alpha 1 Randomized leaky rectified linear unit RReLU f a x a x for x lt 0 x for x 0 displaystyle f alpha x begin cases alpha x amp text for x lt 0 x amp text for x geqslant 0 end cases 3 f a x a for x lt 0 1 for x 0 displaystyle f alpha x begin cases alpha amp text for x lt 0 1 amp text for x geqslant 0 end cases displaystyle infty infty C 0 displaystyle C 0 Tak Tak Ni Exponential linear unit ELU f a x a e x 1 for x lt 0 x for x 0 displaystyle f alpha x begin cases alpha e x 1 amp text for x lt 0 x amp text for x geqslant 0 end cases f a x f a x a for x lt 0 1 for x 0 displaystyle f alpha x begin cases f alpha x alpha amp text for x lt 0 1 amp text for x geqslant 0 end cases a displaystyle alpha infty C 1 when a 1 C 0 otherwise displaystyle begin cases C 1 amp text when alpha 1 C 0 amp text otherwise end cases Tak a 0 displaystyle alpha geqslant 0 Tak 0 a 1 displaystyle 0 leqslant alpha leqslant 1 Tak a 1 displaystyle alpha 1 Scaled exponential linear unit SELU f a x l a e x 1 for x lt 0 x for x 0 displaystyle f alpha x lambda begin cases alpha e x 1 amp text for x lt 0 x amp text for x geqslant 0 end cases z l 1 0507 displaystyle lambda 1 0507 ta a 1 67326 displaystyle alpha 1 67326 f a x l a e x for x lt 0 1 for x 0 displaystyle f alpha x lambda begin cases alpha e x amp text for x lt 0 1 amp text for x geqslant 0 end cases l a displaystyle lambda alpha infty C 0 displaystyle C 0 Tak Ni Ni S shaped rectified linear activation unit SReLU f t l a l t r a r x t l a l x t l for x t l x for t l lt x lt t r t r a r x t r for x t r displaystyle f t l a l t r a r x begin cases t l a l x t l amp text for x leqslant t l x amp text for t l lt x lt t r t r a r x t r amp text for x geqslant t r end cases t l a l t r a r displaystyle t l a l t r a r are parameters f t l a l t r a r x a l for x t l 1 for t l lt x lt t r a r for x t r displaystyle f t l a l t r a r x begin cases a l amp text for x leqslant t l 1 amp text for t l lt x lt t r a r amp text for x geqslant t r end cases displaystyle infty infty C 0 displaystyle C 0 Ni Ni Ni Inverse square root linear unit ISRLU f x x 1 a x 2 for x lt 0 x for x 0 displaystyle f x begin cases frac x sqrt 1 alpha x 2 amp text for x lt 0 x amp text for x geqslant 0 end cases f x 1 1 a x 2 3 for x lt 0 1 for x 0 displaystyle f x begin cases left frac 1 sqrt 1 alpha x 2 right 3 amp text for x lt 0 1 amp text for x geqslant 0 end cases 1 a displaystyle left frac 1 sqrt alpha infty right C 2 displaystyle C 2 Tak Tak Tak Adaptive piecewise linear APL f x max 0 x s 1 S a i s max 0 x b i s displaystyle f x max 0 x sum s 1 S a i s max 0 x b i s f x H x s 1 S a i s H x b i s displaystyle f x H x sum s 1 S a i s H x b i s 4 displaystyle infty infty C 0 displaystyle C 0 Ni Ni Ni SoftPlus f x ln 1 e x displaystyle f x ln 1 e x f x 1 1 e x displaystyle f x frac 1 1 e x 0 displaystyle 0 infty C displaystyle C infty Tak Tak Ni Bent identity f x x 2 1 1 2 x displaystyle f x frac sqrt x 2 1 1 2 x f x x 2 x 2 1 1 displaystyle f x frac x 2 sqrt x 2 1 1 displaystyle infty infty C displaystyle C infty Tak Tak Tak Sigmoid weighted linear unit SiLU a k a Swish f x x s x displaystyle f x x cdot sigma x 5 f x f x s x 1 f x displaystyle f x f x sigma x 1 f x 6 0 28 displaystyle approx 0 28 infty C displaystyle C infty Ni Ni Ni SoftExponential f a x ln 1 a x a a for a lt 0 x for a 0 e a x 1 a a for a gt 0 displaystyle f alpha x begin cases frac ln 1 alpha x alpha alpha amp text for alpha lt 0 x amp text for alpha 0 frac e alpha x 1 alpha alpha amp text for alpha gt 0 end cases f a x 1 1 a a x for a lt 0 e a x for a 0 displaystyle f alpha x begin cases frac 1 1 alpha alpha x amp text for alpha lt 0 e alpha x amp text for alpha geqslant 0 end cases displaystyle infty infty C displaystyle C infty Tak Tak Tak a 0 displaystyle alpha 0 Sinusoyida f x sin x displaystyle f x sin x f x cos x displaystyle f x cos x 1 1 displaystyle 1 1 C displaystyle C infty Ni Ni Tak Sinc f x 1 for x 0 sin x x for x 0 displaystyle f x begin cases 1 amp text for x 0 frac sin x x amp text for x neq 0 end cases f x 0 for x 0 cos x x sin x x 2 for x 0 displaystyle f x begin cases 0 amp text for x 0 frac cos x x frac sin x x 2 amp text for x neq 0 end cases 217234 1 displaystyle approx 217234 1 C displaystyle C infty Ni Ni Ni Gaussian f x e x 2 displaystyle f x e x 2 f x 2 x e x 2 displaystyle f x 2xe x 2 0 1 displaystyle 0 1 C displaystyle C infty Ni Ni Ni Tut H ce funkciya Gevisajda a ye stohastichnoyu zminnoyu vibranoyu z normalnogo rozpodilu pid chas navchannya i zafiksovana yak ochikuvane znachennya rozpodilu do chasu testuvannya Tut s displaystyle sigma logistichna funkciya a gt 0 displaystyle alpha gt 0 vikonuyetsya dlya vsogo intervalu Nastupna tablicya mistit peredavalni funkciyi vid dekilkoh zminnih Nazva Rivnyannya Pohidna ni Oblast Poryadok gladkosti Softmax f i x e x i j 1 J e x j displaystyle f i vec x frac e x i sum j 1 J e x j for i 1 J f i x x j f i x d i j f j x displaystyle frac partial f i vec x partial x j f i vec x delta ij f j vec x 7 0 1 displaystyle 0 1 C displaystyle C infty Maxout f x max i x i displaystyle f vec x max i x i f x j 1 for j argmax i x i 0 for j argmax i x i displaystyle frac partial f partial x j begin cases 1 amp text for j underset i operatorname argmax x i 0 amp text for j neq underset i operatorname argmax x i end cases displaystyle infty infty C 0 displaystyle C 0 Tut d i j displaystyle delta ij simvol Kronekera Div takozhFunkciya vtratPrimitkiKe Lin Du Swamy M N S Neural Networks and Statistical Learning Springer Verlag London 2014 DOI 10 1007 978 1 4471 5571 3 James Keller Derong Liu and David Fogel Fundamentals of computational intelligence neural networks fuzzy systems and evolutionary computation John Wiley and Sons 2016 378 pp ISBN 978 1 110 21434 2 Lionel Tarassenko 2 Mathematical background for neural computing In Guide to Neural Computing Applications Butterworth Heinemann New York 1998 Pages 5 35 ISBN 9780340705896 http doi org 10 1016 B978 034070589 6 50002 6 Anthony Martin 2001 1 Artificial Neural Networks 1 8 doi 10 1137 1 9780898718539 Michael Nielsen Neural Networks and Deep Learning Stegemann J A N R Buenfeld 2014 A Glossary of Basic Neural Network Terminology for Regression Problems Neural Computing amp Applications 8 4 290 296 doi 10 1007 s005210050034 ISSN 0941 0643 Glosarij terminiv z himiyi J Opejda O Shvajka In t fiziko organichnoyi himiyi ta vuglehimiyi im L M Litvinenka NAN Ukrayini Doneckij nacionalnij universitet Doneck Veber 2008 758 s ISBN 978 966 335 206 0 Cybenko G V 2006 Approximation by Superpositions of a Sigmoidal function U van Schuppen Jan H red Springer International s 303 314 Snyman Jan 3 bereznya 2005 Practical Mathematical Optimization An Introduction to Basic Optimization Theory and Classical and New Gradient Based Algorithms Springer Science amp Business Media ISBN 978 0 387 24348 1 Bergstra James Desjardins Guillaume Lamblin Pascal Bengio Yoshua 2009 Departement d Informatique et de Recherche Operationnelle Universite de Montreal Arhiv originalu za 25 veresnya 2018 Glorot Xavier Bengio Yoshua 2010 PDF International Conference on Artificial Intelligence and Statistics AISTATS 10 Society for Artificial Intelligence and Statistics arhiv originalu PDF za 1 kvitnya 2017 Carlile Brad Delamarter Guy Kinney Paul Marti Akiko Whitney Brian 9 listopada 2017 Improving Deep Learning by Inverse Square Root Linear Units ISRLUs arXiv 1710 09967 cs LG Nair Vinod Hinton Geoffrey E 2010 Rectified Linear Units Improve Restricted Boltzmann Machines 27th International Conference on International Conference on Machine Learning ICML 10 USA Omnipress s 807 814 ISBN 9781605589077 Maas Andrew L Hannun Awni Y Ng Andrew Y June 2013 PDF Proc ICML 30 1 Arhiv originalu PDF za 3 sichnya 2017 Procitovano 2 sichnya 2017 He Kaiming Zhang Xiangyu Ren Shaoqing Sun Jian 6 lyutogo 2015 Delving Deep into Rectifiers Surpassing Human Level Performance on ImageNet Classification arXiv 1502 01852 cs CV Xu Bing Wang Naiyan Chen Tianqi Li Mu 4 travnya 2015 Empirical Evaluation of Rectified Activations in Convolutional Network arXiv 1505 00853 cs LG Clevert Djork Arne Unterthiner Thomas Hochreiter Sepp 23 listopada 2015 Fast and Accurate Deep Network Learning by Exponential Linear Units ELUs arXiv 1511 07289 cs LG Klambauer Gunter Unterthiner Thomas Mayr Andreas Hochreiter Sepp 8 chervnya 2017 Self Normalizing Neural Networks arXiv 1706 02515 cs LG Jin Xiaojie Xu Chunyan Feng Jiashi Wei Yunchao Xiong Junjun Yan Shuicheng 22 grudnya 2015 Deep Learning with S shaped Rectified Linear Activation Units arXiv 1512 07030 cs CV Forest Agostinelli Matthew Hoffman Peter Sadowski Pierre Baldi 21 grudnya 2014 Learning Activation Functions to Improve Deep Neural Networks arXiv 1412 6830 cs NE Glorot Xavier Bordes Antoine Bengio Yoshua 2011 Deep sparse rectifier neural networks PDF International Conference on Artificial Intelligence and Statistics Sigmoid Weighted Linear Units for Neural Network Function Approximation in Reinforcement Learning Searching for Activation Functions Godfrey Luke B Gashler Michael S 3 lyutogo 2016 A continuum among logarithmic linear and exponential functions and its potential to improve generalization in neural networks 7th International Joint Conference on Knowledge Discovery Knowledge Engineering and Knowledge Management KDIR 1602 481 486 arXiv 1602 01321 Bibcode 2016arXiv160201321G Gashler Michael S Ashmore Stephen C 9 travnya 2014 Training Deep Fourier Neural Networks To Fit Time Series Data arXiv 1405 2262 cs NE Goodfellow Ian J Warde Farley David Mirza Mehdi Courville Aaron Bengio Yoshua 18 lyutogo 2013 Maxout Networks JMLR WCP 28 3 1319 1327 arXiv 1302 4389 Bibcode 2013arXiv1302 4389G
Топ