InoteveraGaidhi rinotevera
Kuronga Kwechiyero Chekudzidza
Tekinoroji
Nhungamiro yehunyanzvi
Cyclical Learning Rates
Mari yekudzidza yemacyclical inotenderedzazve mwero wekudzidza uchikwira nekudzika pakati pepakati nepamusoro pechisungo pane kungoora chete.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Pa peji ino2 min verenga
Pfupiso
Uku kubhuroka kunopesana kunogona kukurumidzira kusangana uye kunobatsira iyo optimizer kutiza inopinza yemuno minima uye masaddle points.
Kudzika Kwakadzika
Yakakurudzirwa naLeslie Smith muna 2015, cyclical learning rates (CLR) inopikisa fungidziro yekuti mwero unofanira kungodzikira. Pane kudaro, inotenderera pakati pehushoma nepamusoro yakasungwa pamusoro penhamba yakatarwa yekudzokororwa ('kutenderera'), kazhinji iine chimiro chetriangular. Iyo intuition: nguva nenguva kukwidza chiyero kunopa kuputika kwesimba iro rinoita kuti modhi isvetuke kubva muhurombo, inopinza minima uye inoyambuka nzvimbo dzesadhi, nepo dzakaderera dzichiita kuti dzigadzirike. Smith akaunzawo 'LR renji bvunzo' - pfupi pfupi inotsvaira mwero uchikwira uchiona kurasikirwa - kuti uwane miganho yakanaka otomatiki. Triangular, triangular-ne-kuora, uye ine mukurumbira-one-cycle policy zvese zvinovaka pane iyi pfungwa.
Kuitwa muNyika yeChokwadi
fast.ai yakasimudzira mutemo we-one-cycle senzira yekukasira kudzidzisa vagadziri vemifananidzo kusvika pakurongeka kwepamusoro munguva shoma.
Iyo LR renji bvunzo inotsvaira mwero ichikwira pamusoro pemazana mashoma mabheji kuti utore min uye max bounds pamberi pekumhanya chaiko.
Snapshot ensemble inochengetedza modhi yekutarisa pakupera kwekutenderera kwega kwega, ichigadzira yemahara ensemble kubva kune imwe kudzidziswa kumhanya.
Stochastic Gradient Descent ine Warm Restarts (SGDR) nguva nenguva inogadzirisa chiyero kune yakakosha kukosha kutiza inopinza minima.
Ramangwana reKudzidza Kwemabhasikoro
Cyclical schedules uye iyo-one-cycle policy inoramba yakakurumbira kukurumidza kudzidziswa pakuona uye tabular mabasa, uye iyo LR renji bvunzo ijairira tuning trick. Kune mahombe emitauro mamodheru, yakatsetseka yekudziya-pamwe-cosine masheti anowanzo tonga, asi iwo nzwisiso - iyo yehungwaru yekuwedzera inobatsira kutiza nzvimbo dzakaipa dzenzvimbo yekurasikirwa - inozivisa inodziya restarts (SGDR) uye ensemble nzira dzinotora modhi pane imwe neimwe nzvimbo yakaderera. Tarisira kuenderera mberi kwemukume pakati pemazano e cyclical uye anochinja, ega-tuning schedulers.
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 Cyclical Learning Rates 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
Chii chinonzi Cyclical Learning Rates?
Mari yekudzidza yemacyclical inotenderedzazve mwero wekudzidza uchikwira nekudzika pakati pepakati nepamusoro pechisungo pane kungoora chete. Uku kubhuroka kunopesana kunogona kukurumidzira kusangana uye kunobatsira iyo optimizer kutiza inopinza yemuno minima uye masaddle points.
Ndeipi fungidziro huru inodenha cyclical learning rates?
CLR inokwidza nemaune chiyero zvakare uye zvakare, ichiramba maonero echinyakare kuti inofanira kungoora monotonically.
What did Smith report for cyclical learning rates on tested classification tasks?
Smith reported improved classification accuracy without tuning and often in fewer iterations on the tested datasets; the paper does not guarantee a specific loss-landscape escape mechanism.
Ko 'LR range test' inoita sei?
Inomhanya muchidimbu nechiyero chinoramba chichikwira; iyo yekurasika curve inoratidza inonzwisisika hushoma uye huwandu hwekushandisa kutenderera.
Mumutemo we-one-cycle, kukurumidza kunowanzoita sei maererano nechiyero chekudzidza?
Iyo imwe-cycle policy inodzikisira kukurumidza uku iyo chiyero ichikwira uye ichisimudza seyero inodonha, iyo inowedzera muitiro wekugara.
During a triangular cyclical learning-rate cycle, how does the rate move between its bounds?
A triangular policy increases the rate during one half-cycle and decreases it during the other half.
Ramba uchidzidza
Mazano anoenderana
Mamwe madhairekitori akasarudzirwa nyaya iyi