Etalonaasu probabilite
Etalonage dafay tekki ni probabilite yiñ wax ci model bi dañu méngoo ak li am: suñu waxee 70%, xew-xew bi dafa wara am lu tollu ci 70% ci jamono ji.
Résumé
It matters because accurate confidence drives good decisions in medicine, finance, and risk-sensitive AI.
Plongeur bu xóot
Benn model mën na nekk lu jaar yoon waaye kalibre bu baaxul. Reseau yu xóot yi ci jamono jii dañu siiw ci wóolu seen bopp lu ëpp, ñuy génne 99% ci waxtaan yu dëggu yu bariwul luñu koy faral di def. Etalonnage dafay saytu lii ci bucketing waxtaan yi ci wóolu boppam ak xool frequency biñ seetlu ci bucket bu nekk. Diagramu wóor dafay wane li ñu séentu ak li am; model buñ kalibre bu baax mingi toog ci diagonaal bi. Jumtukaayu etalonnage biñ seentu (ECE) dafay tënk bërëb bi ni moyenne pondérée ci biir pañe yi. Fix yi am nañu ñaari cafka: pexe post-hoc yu melni eskaalu Platt (mënngoo ak coppite logistik), eskaalu tàngoor (seddale logits ak T scalar buñ jàng), ak régression isotonic (jëfandikoo jéego bu monotonik); ak pexe ci diiru tàggat yaram yu melni etiketu nooy wala ñàkka am poñ yu jaar yoon. Etalonnage ak jaar yoon ñaari mébet yu wuute lañu, te gëna suqali benn ci ñoom jarul gëna suqali beneen bi.
Gis-gis xarala
Eskalaasioŋ tàngoor mooy li gëna am solo ci reso neuronal yi: xaaj logits pre-softmax ak benn tàngoor T buñ jàng, ba noppi defaraat softmax. T > 1 dafay yombal distribution yu gëna wóolu seen bopp, T < 1 dafay ñaw leen. Li gëna am solo mooy T mën na nekk ci done yu baax ngir wàññi log-likelihood bu baaxul te du musa soppi ban class mooy jël ndam li, kon njubte gi du laal fekk probabilite yi dañuy nekk njub. Benn parametre bi moo tax mu am done yu baax te daanaka mënu ñu ko def lu ëpp li war.
njeextalu pexe
Njëgg ak budget
Dogal yi architecture di jël dañuy indi njariñ ak njëgu liggéey bi ay at ci ginaaw.
dogal yu gëna leer
Njàngalem xarala yi dafay jàppale ekip yi ñu tànn li gën, te baña yam ci li gëna bees daal.
Xool kalite
Tanneef yu gëna baax ci wàllu ingeñër dina wàññi jafe-jafe yi ci wàllu wóor ci liggéey bi.
Ëlëgu etalonnage probabilite
Bi IA duggee ci ay loop yu am solo, etalonnage bi dafay joge ci xalaat ginaaw dem ci laaj bi. Liggéey bi dafay gëna yaatu ngir etalonnage xeetu làkk bu yaatu wóolu boppam ak ñàkka wóor buñ wax, etalonnage ci suufu coppite séddale, ak etalonnage ci grupu-wise suko defee probabilite yi gëna jub ci subpopulations yi. Xaarandi metrics etalonnage ci wetu njub ci kàrtu model ak audit yuñ yamale, boole ci boole bu gëna dëgër ak waxtaan bu méngoo ak waxtaan bu tànneef suko defee sistem yi mëna baña gëm su seen wóolu seen bopp wàññeeku.
Doxal ci àdduna dëgg
Ab serwiisu meteo dafay fexe matal bis yi ñu wax ni dina am 30% taw ci dëgg taw lu tollu ci 30% ci jamono ji, muy mébetu etalonnage bu téere bi.
Xeetu leble-default dafay aju ci tàngoor, kon risk default bu 5% biñu wax mingi méngoo ak 5% ci njëgu leble yi.
Reseau diagnostic médical dañu koy kalibrewaat ak regression isotonic suko defee 'probabilite bu rëy ci feebar' dafay wane dëgg laata doktër yi di jëfandikoo.
Benn pilu gis-gis buy dawal boppam mooy kalibre wóolu gis-gis mbir suko defee 90% ci poñ yi doxkat yi am dañu ko wóolu ci anam wu jaar yoon ci modulu waajal bi.
Risk yi ak balustrade yi
Optimize benn benchmark mën na nëbb ñakk kattan yu gëna yaatu ci sistem bi.
Njëg li ñuy fay ci infrastructure yi ak ci toppatoo dañuy faral di suufeel.
Bu sistem yi di gëna xawa jafee xam, jafe-jafe yi am ci wàllu kaaraange ak seetlu mën nañu gëna bari.
Roadmap ngir samp gi
Mandargal latency, kalite, ak njëg yi laata ngay jëfandikoo.
Benchmark ci biir sargal ak done yu dëggu.
Jumtukaay bi di saytu njuumte yi, derive bi ak njeextalu jëfandikukat bi.
Waajal rollback ak yooni tontu ci jafe-jafe yi laata ngay eskale.
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 Probability Calibration 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
Etalonaasu wóolu IA
Laaj yi ñuy faral di laaj
What is Probability Calibration?
Etalonage dafay tekki ni probabilite yiñ wax ci model bi dañu méngoo ak li am: suñu waxee 70%, xew-xew bi dafa wara am lu tollu ci 70% ci jamono ji. Dafa am solo ndax wóolu sa bopp mooy tax nga jël dogal yu baax ci wàllu medsin, xaalis ak IA bu bari risk.
Lan la tàngoor di def ci génnug reso neuronal bi?
Eskalaasioŋ tàngoor dafay xaaj logits ak benn T buñ jàng ba noppi jëfandikoowaat softmax, nooyloo wala ñawloo probabilite yi te du soppi argmax bi.
Luy diagramu wóor?
Semaawu wóolu dafay méngale wóolu biñ séentu ak limuy am ci njariñ li; diagonaal bi dafay màndargaal etalonnage bu jaar yoon.
Lan moo waral model bu jaar yoon mën na soxla etalonnage?
Benn model mën na tànn li ñuy wax luy waaja am (ci dëggu bu rëy) waaye mën na joxe valeur probabilite yu wuute, kon dañu wara xool etalonnage bi bu baax.