Nhazi nke puru omume
Nhazi pụtara ihe omume nke ihe nlereanya kwuru dabara na eziokwu: mgbe ọ na-ekwu 70%, ihe omume kwesịrị ime ihe dịka 70% nke oge.
Nchịkọta
It matters because accurate confidence drives good decisions in medicine, finance, and risk-sensitive AI.
Ime miri emi
Ụdị nwere ike ịbụ nke ziri ezi ma ọ dị mma nke ọma. Netwọk miri emi nke oge a bụ ihe ama ama maka oke ntụkwasị obi, na-ewepụta 99% na amụma na-adịchaghị obere oge. Calibration na-enyocha nke a site na ịbu amụma site na ntụkwasị obi na ịlele ugboro ole a hụrụ na bọket ọ bụla. E buru amụma na eserese a pụrụ ịdabere na ya ka ọ dị adị; ihe nlere anya nke ọma na-anọdụ na diagonal. Njehie calibration a tụrụ anya ya (ECE) na-achịkọta oghere ahụ dị ka nkezi dị arọ gafere bins. Ndozi na-abịa na ekpomeekpo abụọ: ụzọ post-hoc dị ka Platt scaling (na-adaba na mgbanwe logistic), ikpo ọkụ okpomọkụ (na-ekewa logits site na scalar T mụtara), na isotonic regression (a monotonic step fit); na ụzọ ọzụzụ oge dị ka akara mbelata ma ọ bụ mfu akara kwesịrị ekwesị. Nhazi na izi ezi bụ ihe mgbaru ọsọ dị iche iche, na imelite otu ọ dịghị mkpa ka ọ ka mma nke ọzọ.
Nghọta nka nka
Mkpokọta okpomoku bụ ihe na-arụ ọrụ maka ụgbụ akwara: kesaa logits pre-softmax site na otu okpomọkụ mụtara T, wee re-softmax. T > 1 na-eme ka nkesa nwere ntụkwasị obi dị nro, T <1 na-amụba ha. N'ụzọ dị mkpa T dabara na data nkwado iji belata ohere ndekọ na-adịghị mma ọ dịghị agbanwe agbanwe nke klaasị meriri, yabụ emetụghị izi ezi ebe ohere ga-abụ eziokwu. Otu paramita ya na-eme ka ọ na-arụ ọrụ data na ọ fọrọ nke nta ka ọ bụrụ ihe na-agaghị ekwe omume ịfefe.
Mmetụta atụmatụ
Ọnụ ego na mmefu ego
Mkpebi ihe owuwu ụlọ na-akwalite arụmọrụ yana ọnụ ahịa ọrụ ruo ọtụtụ afọ.
Mkpebi doro anya
Nkà mmụta nka na-enyere ndị otu egwuregwu aka ịhọrọ nchịkọta ziri ezi, ọ bụghị naanị nke kachasị ọhụrụ.
Quality akara
Nhọrọ injinia ka mma na-ebelata ihe omume ntụkwasị obi na mmepụta.
Ọdịnihu nke nhazi nke puru omume
Ka AI na-abanye na loops dị elu, nhazigharị na-esi n'echiche emechaa gaa na ihe achọrọ. Ọrụ na-agbasawanye n'ịkwado ntụkwasị obi ụdị asụsụ buru ibu yana enweghị nkwuwa okwu ejighị n'aka, nhazigharị n'okpuru ngbanwe nkesa, yana nhazi nke amamihe dị n'otu ka ihe puru omume dị mma n'ofe obodo nta. Na-atụ anya metrik nzizi n'akụkụ izi ezi na kaadị ihe nlereanya yana nyocha usoro, gbakwunyere njikọta siri ike na amụma dabara adaba na amụma nhọrọ ka sistemụ wee nwee ike jụ ntụkwasị obi mgbe ntụkwasị obi eziokwu ha dị ala.
Mmejuputa n'ezie n'ụwa
Ọrụ ihu igwe na-achọpụta na amụma ụbọchị na 30% mmiri ozuzo na-ahụ n'ezie mmiri ozuzo ihe dịka 30% nke oge, ebumnuche nhazi akwụkwọ ọgụgụ.
Ụdị ndabara kredit bụ nke nwere oke okpomọkụ, yabụ, ihe egwu ndabara 5% dabara adaba n'ezie na ọnụego ndabara akụkọ ihe mere eme 5% maka mbinye ọnụahịa.
A na-emegharị netwọk nchoputa ahụike site na isotonic regression ka 'oke puru nke ọrịa' na-egosipụta ezi ọnọdụ tupu ndị dọkịta emee ihe.
Otu nleba anya nke ịnya ụgbọ ala na-emezi ntụkwasị obi nchọta ihe ka ejiri modul atụmatụ tụkwasara akara ndị na-agafe agafe 90% nke ọma.
Ihe ize ndụ & okporo ụzọ nche
Ịkwalite otu akara ngosi nwere ike zoo adịghị ike sistemụ sara mbara.
A na-eledakarị ihe akụrụngwa na ụgwọ ọrụ anya.
Ọdịiche nchekwa na nleba anya nwere ike itolite ka sistemu na-adịwanye mgbagwoju anya.
Map mmejuputa
Kọwaa latency, ịdịmma na ebumnuche ọnụ ahịa tupu mmejuputa ya.
Benchmark n'okpuru ibu dị adị na ọnọdụ data.
Nleba anya akụrụngwa maka mperi, ịkpafu na mmetụta onye ọrụ.
Kwadebe ụzọ nzaghachi azụghachi azụ na ihe omume tupu ịchachaa.
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 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
Ntuziaka na-esote
AI Trust Calibration
Ajụjụ a na-ajụkarị
What is Probability Calibration?
Nhazi pụtara ihe omume nke ihe nlereanya kwuru dabara na eziokwu: mgbe ọ na-ekwu 70%, ihe omume kwesịrị ime ihe dịka 70% nke oge. Ọ dị mkpa n'ihi na ntụkwasị obi ziri ezi na-akpali mkpebi dị mma na ọgwụ, ego, na AI nwere ihe egwu.
Kedu ihe mbelata okpomọkụ na-eme na mpụta netwọkụ akwara ozi?
Ọkpụkpọ okpomoku na-ekewa logit site na otu T mụtara wee tinyegharịa softmax, ime ka ọ dị nro ma ọ bụ na-eme ka ọ dị nkọ na-agbanweghị argmax.
Kedu ihe eserese ntụkwasị obi na-akpa nkata?
Eserese a pụrụ ịdabere na ya na-atụnyere ntụkwasị obi e buru amụma na ugboro ole ọ ga-esi na ya pụta; diagonal na-anọchi anya nhazigharị zuru oke.
Kedu ihe kpatara ihe nlere nke ziri ezi ka ji chọọ nhazigharị?
Ihe nlereanya nwere ike ịdo amụma nke ọma (oke izi ezi) ma na-ekenye ụkpụrụ puru omume ekwekọghị, yabụ a ga-enyocharịrị iche iche.