Selụ ebe nchekwa ogologo oge dị mkpirikpi
Selụ ebe nchekwa ogologo oge (LSTM) bụ ụdị pụrụ iche nke ngalaba netwọkụ akwara na-emegharị ugboro ugboro wuru iji cheta ozi n'ofe ogologo usoro.
Nchịkọta
They solved the vanishing-gradient problem that crippled earlier RNNs, powering a decade of breakthroughs in language, speech, and translation.
Ime miri emi
Ndị Sepp Hochreiter na Jurgen Schmidhuber webatara na 1997, cell LSTM na-edobe 'steeti cell' nke na-eme dị ka eriri ebe nchekwa na-aga n'usoro. Ọnụ ụzọ ámá atọ mụtara na-achịkwa ya: ọnụ ụzọ nchefu na-ekpebi ihe a ga-ehichapụ, ọnụ ụzọ ntinye na-ekpebi ozi ọhụrụ a ga-echekwa, ọnụ ụzọ mmepụta na-ekpebi ihe ọ ga-ekpughe dị ka mmepụta cell. Ọnụ ụzọ ámá ọ bụla na-eji sigmoid (mpụta 0 ruo 1) mee ihe dị ka ngbanwe dị nro. N'ihi na a na-emelite steeti cell nke ukwuu site na mgbakwunye kama ịba ụba ugboro ugboro, gradients nwere ike na-aga azụ n'ọtụtụ usoro oge na-eme ka ọ ghara ịdaba na efu, na-ahapụ LSTM ka ịmụta dabere na narị nzọụkwụ dị iche iche. Tupu Transformers, LSTM kwadoro Google Ntụgharị asụsụ, njirimara okwu, na ọgbọ ederede.
Nghọta nka nka
Ndozi nke na-apụ n'anya-gradient na-abịa site na mmelite nso nso linear steeti: c_t = f_t * c_{t-1} + i_t * g_t. Ọnụ ụzọ chefuo f_t (sigmoid) nwere ike ịnọ nso 1, na-emepụta 'carousel njehie na-adịgide adịgide' ka akara njehie na-adị ndụ n'azụ-site n'oge gafere ogologo oge. Ọnụ ụzọ ámá bụ onwe ha obere oghere akwara (sigmoid maka gating, tanh maka ụkpụrụ ndị ndoro-ndoro anya), ha niile zụrụ azụ site na mgbada gradient. Nke a gating na-eme ka netwọk mụta ihe ị ga-edobe na ihe ị ga-atụfu.
Mmetụta atụmatụ
Mkpebi doro anya
Ọ na-enyere gị aka ikewapụta nkwupụta ọrụ aka doro anya na asụsụ ahịa.
Ọnụ ego na mmefu ego
Ị nwere ike ịjụ ajụjụ mmejuputa iwu ka mma tupu itinye ego ma ọ bụ oge.
Team na usoro ọrụ
Ndị otu nwere nghọta na-eme ka ngwaahịa, amụma na mkpebi mmụta ka mma.
Ọdịnihu nke mkpụrụ ndụ ebe nchekwa ogologo oge dị mkpirikpi
Ndị na-eme mgbanwe agafeela LSTM maka nnukwu ọrụ asụsụ n'ihi na ha na-emekọ ihe n'usoro wee na-eji nlebara anya na-eme ogologo oge, ebe usoro LSTM na-egosi otu nzọụkwụ n'otu oge. N'agbanyeghị nke ahụ, LSTMs ka bara uru maka mgbasa ozi, obere oge, yana ntọala nwere ikike, yana na data usoro oge dị ntakịrị. Ọrụ na-adịbeghị anya dị ka xLSTM (2024) na-elegharị anya ma na-emeziwanye ihe owuwu ahụ site na gating ọhụrụ na ebe nchekwa iji mpi n'ọtụtụ, na-egosi na agwụchabeghị echiche ahụ.
Mmejuputa n'ezie n'ụwa
Ntụgharị asụsụ igwe na-enye ike n'isi mmalite Google Sistemụ akwara nke ntụgharị tupu Transformers eweghara.
Nchọpụta okwu-gaa-ederede na ndị enyemaka olu na ngwa nkwuwa okwu.
Na-ebu amụma ụkpụrụ ga-eme n'ọdịnihu na usoro oge dị ka ọchịchọ ike, ọgụgụ ihe mmetụta, ma ọ bụ ọnụ ahịa ngwaahịa.
Na-emepụta ederede ma ọ bụ egwu otu akara n'otu oge yana ịmecha usoro.
Ihe ize ndụ & okporo ụzọ nche
Otu dị iche iche nwere ike iji otu okwu ahụ mee ihe n'ụzọ dị iche, yabụ kọwapụta oge n'oge.
Ihe nrịbama nwere ike ịdị ike ebe arụmọrụ ụwa na-adaghị adaba.
Ileghara ogo data na atụmatụ nyocha anya na-emepụtakarị nsonaazụ na-adịghị mma.
Map mmejuputa
Malite na nkọwa asụsụ dị larịị nke nsonaazụ ịchọrọ.
Họrọ otu metrik ịga nke ọma na otu ọnọdụ ọdịda tupu nnwale.
Gbaa obere onye na-anya ụgbọ elu nwere data nnọchite anya, ọ bụghị ihe ngosi ngosi na-egbu maramara.
Detuo ebe Selụ ebe nchekwa ogologo oge na-enyere aka yana ebe ụzọ dị mfe ka mma.
Nọgide na-eme nchọpụta
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Long Short-Term Memory Cells quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ntuziaka na-esote
Njikwa ebe nchekwa GPU na nkewa
Ajụjụ a na-ajụkarị
What is Long Short-Term Memory Cells?
Selụ ebe nchekwa ogologo oge (LSTM) bụ ụdị pụrụ iche nke ngalaba netwọkụ akwara na-emegharị ugboro ugboro wuru iji cheta ozi n'ofe ogologo usoro. Ha doziri nsogbu na-apụ n'anya-gradient nke mebiri RNN mbụ, na-eme ka afọ iri nwee ọganihu n'asụsụ, okwu, na ntụgharị asụsụ.
Kedu ọnụ ụzọ ámá atọ na-achịkwa ozi na-eru na cell LSTM ọkọlọtọ?
Otu LSTM na-eji ọnụ ụzọ nchefu (ihe a ga-ehichapụ), ọnụ ụzọ ntinye (ihe a ga-echekwa), yana ọnụ ụzọ mmepụta (ihe a ga-ekpughe), nke ọ bụla sigmoid mụtara.
Kedu nnukwu nsogbu dị na RNN mbụ LSTM doziri?
Ọkọlọtọ RNNs na-ata ahụhụ gradients na-apụ n'anya nke na-egbochi ịmụta ịdabere ogologo oge; steeti cell mgbakwunye LSTM na-ahapụ gradients ka ọ nọgide.
Ònye webatara LSTM, na kedu afọ?
Sepp Hochreiter na Jurgen Schmidhuber bipụtara LSTM na 1997.
Kedu ihe kpatara steeti cell LSTM ji enyere gradients aka ịlanarị ọtụtụ usoro oge?
Mmelite mgbakwunye ahụ ('carousel njehie mgbe niile') na-ezere ịba ụba ugboro ugboro na-ebelata gradients, yabụ akara njehie na-alaghachi azụ n'ogologo.
Kedu ọrụ mmeghari ka ọnụ ụzọ LSTM na-ejikarị eme ihe dị ka mgbanyụ ma ọ bụ gbanyụọ nro?
Gates na-eji sigmoid, onye mmepụta 0-to-1 nwere ọnụọgụ ozi ole na-agafe, na-eme dị ka mgba ọkụ dị nro.