Selul yuy fàttaliku lu yàgg te gàtt
Long Short-Term Memory (LSTM) selul yi xeetu reso neuronal lañu yuñ tabax ngir fàttaliku xibaar yi ci diir bu yàgg.
Résumé
They solved the vanishing-gradient problem that crippled earlier RNNs, powering a decade of breakthroughs in language, speech, and translation.
Plongeur bu xóot
Sepp Hochreiter ak Jurgen Schmidhuber ñoo ko dugal ci atum 1997, selil LSTM dafay tëye 'statu selil' buy def ni sentiir buy yóbbu fàttaliku biy daw ci biir toppalante bi. Ñatti buntu yuñ jàng ñoo koy yoree: buntu fàtte bi mooy xool li ñu wara efaase, buntu dugal bi mooy xool ban leeral bu bees lañu wara denc, buntu genn bi mooy xool li ñu wara fësal muy génnug selil bi. Gate bu nekk dafay jëfandikoo sigmoid (génne 0 ba 1) ngir nekk commutateur bu woyof. Ndax tolluwaayu selil bi dañu koy gëna yeesal ci yokk, du bariwaat, gradient yi mën nañu dellu ginaaw ci diir bu bari te duñu wàññeeku ba amul dara, loolu mooy tax LSTM yi jàng dependency yu téemeeri jéego yu wuute. Laata Transformers yi, LSTM yi ñoo daan jàppale Tekki làkk, xàmmee kàddu yi ak defar mbind.
Gis-gis xarala
Fix gradient biy réer mingi aju ci yeesali selil bi ci anam wu jege ligneer: c_t = f_t * c_{t-1} + i_t * g_t. Buntu fàtte f_t (sigmoid) mën na des ci wetu 1, sos 'carousel njuumte bu sax' suko defee siñaal njuumte yi mëna dundu backpropagation-ci-diir ci diir bu yàgg. Buntu yi ci seen bopp ay couche neuronal yu ndaw lañu (sigmoid ngir gating, tanh ngir valeur candidat), ñoom ñépp ñu ngi leen tàggat ci wàccinu gradient. Gating bii dafay tax reso bi xam limu wara denc ak limu wara sànni.
njeextalu pexe
dogal yu gëna leer
Daf lay jàppale nga tàqale kàddu yu leer ci wàllu xarala ak làkku fësal njaay.
Njëgg ak budget
Mën nga laaj laaj yu gëna baax ci samp gi balaa ngay dugal xaalis wala sa jotu liggéey.
Ekip ak def liggéey
Ekip yi bokk xam-xam ñoo gëna mëna jël yenn dogal ci wàllu produit, politik ak jàng.
Ëlëgu selil yu am mémoire yu gudd te gàtt
Transformers yi dañu romb LSTMs ci liggéey yu yaatu ci làkk ndax dañuy parallel ci benn toppalante ba noppi jàpp contexte bu sori jaaraleko ci bàyyi xel, waaye LSTMs yi dañuy liggéey token yi benn jéego ci yoon. Ba leegi, LSTMs ñu ngi wéy di am solo ci streaming, latency bu woyof, ak jekkal yu am jumtukaay yu néew, ak ci done yu ndaw yu am ci jamono. Liggéey bu bees bu melni xLSTM (2024) dafa xoolaat ak beesal architecture bi ak gating ak mémoire bu bees ngir mëna joŋante ci scale, di wane ni xalaat bi jeexagul.
Doxal ci àdduna dëgg
Tekkikatu masin buy dooleel sistem neuronal bu Tekkikatu làkk balaa Transformers di jël liggéey bi.
Xamme kàddu ci bind ci assistant baat ak losisel buy bind.
Xalaatal valeur yu ëlëg ci diir bu gàtt lu ci melni laaj energie, li kaptër yi di jàng, wala njëgi aksioŋ yi.
Defar mbind wala music benn token benn yoon ak toppalante ci boppam.
Risk yi ak balustrade yi
Ekip yu bari mën nañu jëfandikoo benn baat ci anam wu wuute, kon teela leeral yaatuwaayam.
Benchmark yi mën nañu nuru lu am doole waaye performance yi ci àdduna bi duñu tolloo.
Bëgg kalite done ak palaŋu jàngat dafay faral di jur njariñ yu yomba dagg.
Roadmap ngir samp gi
Tàmbaleel ci joxe leeral ci làkk wu leer ci njariñ li nga soxla.
Tannal benn metric bu baax ak benn anam bu baaxul balaa ngay saytu.
Doxal ab pilote bu ndaw ak ay done yu representatif, du ab demo bu leer.
Dokumenteer fi selul yu am memory bu gudd te gàtt di jàppale ak fi pexe yu gëna yomba gëna baax.
Weyal di banneexu
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
Gis bi ci topp
Doxal mémoire GPU ak xaaj
Laaj yi ñuy faral di laaj
What is Long Short-Term Memory Cells?
Long Short-Term Memory (LSTM) selul yi xeetu reso neuronal lañu yuñ tabax ngir fàttaliku xibaar yi ci diir bu yàgg. Ñu saafara jafe-jafe vanishing-gradient bi daan gàllankoor RNN yu njëkk ya, loolu moo jur jéego yu am solo ci làkk, wax ak tekki làkk ci diiru fukki at.
Yan ci ñetti buntu yi ñooy saytu ni xibaar bi di jaar ci selil LSTM biñ miin?
LSTM dafay jëfandikoo buntu fàtte (li ñu wara efaase), buntu duggal (li ñu wara denc), ak buntu genn (li ñu wara fësal), bu nekk sigmoid buñ jàng.
Lan mooy jafe-jafe bi gëna mag ci RNN yu njëkk ya LSTM yi saafara?
RNN yiñ miin dañuy am gradient yuy jeex yuy tere ñu jàng dependence yu sori; tolluwaayu selil biy yokk ci LSTM dafay tax gradient yi wéy.
Kan moo dugal LSTM, ak ci ban at?
Sepp Hochreiter ak Jurgen Schmiduber ñoo siiwal LSTM ci atum 1997.
Lan moo waral nekkinu selil LSTM di jàppale gradient yi ñu mëna dundu ci diir bu yàgg?
Coppite biy yokk ('carousel njuumte bu sax') dafay moytu bariwaay yu bari yuy wàññi gradient yi, suko defee siñaal njuumte yi di dellu ci diir bu yàgg.
Ban fonction activation la buntu LSTM yi di faral di jëfandikoo ngir nekk interrupteur on/off yu woyof?
Gates dafay jëfandikoo sigmoid, muy génne 0-ba-1 ngir xayma bariwaayu xibaar biy jaar, muy def ni commutateur bu woyof.