Machine learningTime-series forecasting

Reformer: Efikasan Transformer za duge sekvence

Reformer je efikasna varijanta Transformer arhitekture koju su predstavili Kitaev, Kaiser i Levskaya na ICLR 2020. On rešava problem prevelike memorijske i računarske složenosti standardne samo-pažnje, koja iznosi O(L²), za duge sekvence. Ključne inovacije su pažnja zasnovana na heširanju osetljivom na lokalitet (LSH), koja aproksimira punu pažnju u O(L log L) vremenu, i reverzibilni rezidualni slojevi koji dramatično smanjuju memoriju aktivacija tokom treninga.

Otvorite u MethodMindUskoroVideoUskoroDownload slides

Pročitajte celu metodu

Samo za članove

Prijavite se besplatnim nalogom da biste pročitali ovaj odeljak.

Prijavite se

Method map

The neighbourhood of related methods — select a node to explore.

Reformer: Efikasan Transformer za duge sekvence
InformerPyraformer

Izvori

  1. Kitaev, N., Kaiser, Ł., & Levskaya, A. (2020). Reformer: The efficient transformer. ICLR. link

Kako citirati ovu stranicu

ScholarGate. (2026, June 2). Reformer (The Efficient Transformer). ScholarGate. https://scholargate.app/sr/deep-learning/reformer

Which method?

Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.

Compare side by side

Citirana u

ScholarGateReformer (Reformer (The Efficient Transformer)). Preuzeto 2026-06-15 sa https://scholargate.app/sr/deep-learning/reformer · Skup podataka: https://doi.org/10.5281/zenodo.20539026