InoteveraGaidhi rinotevera
Cyclical Learning Rates
Tekinoroji
Nhungamiro yehunyanzvi
Kuronga Kwechiyero Chekudzidza
Chirongwa chechiyero chekudzidza chinoshandura saizi yenhanho panguva yekudzidziswa pane kuibata yakagadziriswa.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Pa peji ino2 min verenga
Pfupiso
Getting it right is often the single biggest lever for whether a model converges quickly and reaches high accuracy.
Kudzika Kwakadzika
Mwero wekudzidza unodzora kuti yakakura sei nhanho iyo optimizer inotora yega yega update. Kunyanya kukwira uye kudzidziswa kunosiyana; yakaderera uye inokambaira kana kunamira. Kuronga kunogadzirisa kukosha uku nekufamba kwenguva. Mabikirwo akajairika emazuva ano iwarmup inoteverwa nekuora: tanga pedyo ne zero uye ukwire kumusoro kwemazana mashoma ekutanga kana zviuru nhanho (nekutanga, magiredhi ane ruzha haaputire huremu husina kugadzikana), wozodzikira zvishoma nezvishoma. Maumbirwo akakurumbira ekuora anosanganisira kuora kwenhanho (kudonha nechinhu panguva yakatarwa), kuora kweexponential, uye cosine annealing, iyo inotevedza hafu-cosine curve pasi kusvika pedyo ne zero. Zvirongwa zveCosine zvine mutsara wekudziya zvino zvave chiyero chekudzidzisa mamodheru emitauro mikuru, nepo cyclical uye imwe-kutenderera mitemo inogona kukurumidzira diki-modhi kudzidziswa.
Kuitwa muNyika yeChokwadi
Linear warmup plus cosine decay inoshandiswa pakugadzirira mamodheru emutauro wetransformer.
Kuora kwenhanho kunodonhedza mwero wekudzidza 10x panguva dze30, 60, uye 90 paunenge uchidzidzisa vagadziri vemifananidzo paImageNet.
Iyo ye-one-cycle policy in fast.ai kudzidzisa modhi kune yakanaka yekurongeka mune mashoma epochs.
Cosine annealing ine inodziya restarts kuti nguva nenguva kutiza kurasikirwa kwakapinza minima uye kunatsiridza generalization.
Ramangwana Rekuronga Mwero weKudzidza
Sezvo kumhanya kwekudzidzira kuchiwedzera kudhura, mashedhi ari kugadzirwa pamwe chete nema optimizers uye saizi yebatch, uye vaongorori vanodzidza mitemo yekuyera kufanotaura mwero wepamusoro wepamusoro usati wadzidziswa. Schedule-isina optimizers inobvisa kudiwa kwekutora kuora curve pachine nguva iri kuwana traction, uye adaptive, mhinduro-inotungamirwa masheti anopindura kuhupenyu kurasikirwa macurves anogona kuderedza kuyedzwa-uye-kukanganisa kuchiri kutonga kudzidziswa kwakakura.
Kuzvishandisa
Nei zvichikosha
- Mutengo uye bhajeti. Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore.
- Sarudzo dzakajeka. Dzidzo yehunyanzvi inobatsira zvikwata kusarudza murwi wakakodzera, kwete iwo mutsva chete.
- Kudzora kwemhando yepamusoro. Sarudzo dzeinjiniya dziri nani dzinoderedza zviitiko zvekuvimbika mukugadzira.
Chenjera
- Kugadzirisa imwe bhenji kunogona kuvanza yakafara system kushaya simba.
- Infrastructure uye mari yekugadzirisa inowanzotarisirwa pasi.
- Chengetedzo uye kucherechedzwa mapundu anogona kukura sezvo masisitimu anowedzera kuoma.
Matanho anotevera
- Tsanangura latency, mhando, uye mutengo zvinangwa usati waitwa.
- Benchmark pasi pechokwadi mutoro uye data mamiriro.
- Chishandiso chekutarisa zvikanganiso, kudonha, uye mushandisi maitiro.
- Gadzirira nzira dzekudzosera kumashure uye dzezviitiko usati wawedzera.
Ramba Uchiongorora
Tsamba yemahara
Tora iyo yemazuva ese AI pfupiso
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Take the Learning Rate Scheduling quiz
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
Mibvunzo inowanzo bvunzwa
What is Learning Rate Scheduling?
Chirongwa chechiyero chekudzidza chinoshandura saizi yenhanho panguva yekudzidziswa pane kuibata yakagadziriswa. Kuigadzirisa kazhinji ndiyo imwe hombe lever yekuti modhi inochinjika nekukurumidza uye inosvika pakurongeka kwepamusoro.
Chii chinangwa chechikamu che'warmup' muchirongwa chechiyero chekudzidza?
Warmup inotanga pedyo ne zero uye inowedzera muyero kuitira kuti kusagadzikana kwekutanga zvigadziriso zvisavhiringa modhi iyo nhamba yeoptimizer isati yagadzikana.
Ndeipi purogiramu inonyatso tevera hafu-cosine curve pasi kusvika pachiyero chidiki?
Cosine annealing inoodza mwero wekudzidza pamwe nehafu-cosine chimiro, ichipa nhanho huru dzekutanga uye nhanho diki pedyo nekumagumo.
Chii chinoitika kana mwero wekudzidza ukakwidzwa zvakanyanya?
Chiyero chakanyanya kukwira chinokonzeresa kuwanda, saka kurasikirwa kunogona kutenderera kana kuputika pane kugadzirisa.
Kuora kwedanho kunoitei pachiyero chekudzidza?
Kuora kwenhanho kunowedzera mwero nechinhu (semuenzaniso, 0.1) pamatanho akasarudzwa, kugadzira patani yemasitepisi.
What does a warm restart do to a learning-rate schedule?
Warm restarts begin another learning-rate cycle at a higher value; they do not guarantee escape from a particular loss-landscape feature.
Ramba uchidzidza
Mazano anoenderana
Mamwe madhairekitori akasarudzirwa nyaya iyi