Asarar Sau uku da Koyon Ma'auni
Hasara sau uku tana koyar da hanyar sadarwa ta jijiyoyi don sanya abubuwa iri ɗaya kusa da juna da kuma abubuwa iri ɗaya nesa ba kusa ba a cikin sararin samaniya.
Dubawa
It is the foundation behind face recognition, image search, and recommendation systems that need to compare things rather than just classify them.
Zurfafa nutsewa
Koyon awo yana horar da abin ƙira don samar da abubuwan haɗawa, ɓangarori inda nisa ke nuna kamanni. Rashin sau uku yana yin haka ta amfani da bayanai guda uku a lokaci guda: anga, tabbatacce (aji ɗaya da anka), da mara kyau (aji daban-daban). Manufar tana tura anka kusa da tabbatacce fiye da mara kyau ta aƙalla ƙayyadadden gefe. A bisa ka'ida, asarar ita ce max (0, d (a,p) - d(a,n) + gefe), inda d yawanci nisa ne na Euclidean. Google's 2015 FaceNet ya shahara da wannan tsarin, yana koyan abubuwan haɗa fuska mai girma 128 kai tsaye. Da zarar an horar da ku, kuna kwatanta kowane abu biyu ta hanyar ƙididdige nisa, babu sake horon da ake buƙata don sababbin ainihi. Wannan damar buɗe-ido shine dalilin da ya sa ma'auni koyo ikon tabbatarwa da rarrabuwa ayyuka na dawo da su ba zai iya sauƙin ɗauka ba.
Fahimtar Fasaha
Gefen ita ce ke sa asarar sau uku ta yi aiki. Idan ba tare da shi ba, samfurin zai iya rushe duk abubuwan da aka haɗa da su zuwa wuri guda, wanda zai sa kowane nisa ya zama sifili kuma tsari mara ma'ana. Margin yana tilasta majigi: dole ne mara kyau ya kasance aƙalla tazara fiye da tabbatacce kafin asarar ta kai sifili. Abubuwan haɗawa yawanci L2-na al'ada a kan naúrar hypersphere, don haka nisa yana da iyaka da kamanta. Zaɓin gefe (sau da yawa a kusa da 0.2) ya bambanta yadda azuzuwan suka taru akan rabuwa tsakanin su.
Dabarun Tasiri
Shawarwari masu haske
Yana taimaka muku keɓance bayyanannen da'awar fasaha daga harshen talla.
Kudin da kasafin kuɗi
Kuna iya yin mafi kyawun tambayoyin aiwatarwa kafin kashe kuɗi ko lokaci.
Ƙungiya da aikin aiki
Ƙungiyoyin da ke da fahimtar juna suna yin mafi kyawun samfura, manufofi, da yanke shawara na koyo.
Makomar Asarar Sau Uku da Koyon Ma'auni
Ana ƙara maye gurbin tsaftataccen asarar sau uku da maƙasudai masu faɗi kamar kamanceceniya da yawa, wakili-anga, da asarar bambanci (InfoNCE) waɗanda ke kwatanta nau'i-nau'i da yawa a kowane mataki kuma suna haɗuwa cikin sauri. Hanyoyin kulawa da kai irin su SimCLR suna nuna ilimin awo na iya aiki ba tare da lakabi ba ta hanyar ɗaukar ra'ayoyi masu ƙarfi a matsayin tabbatacce. Kamar yadda ma'ajin bayanai na vector da haɓaka-dawowa-ƙaramar haɓakar tsarawa, koyaswar haɗaɗɗiya ke ƙarfafa bincike na ma'ana a sikelin abubuwa biliyan, don haka ainihin ra'ayin nesa-kamar- kamanceceniya yana zama mafi tsakiya, ko da ƙayyadadden tsari na uku-uku yana dushewa.
Aiwatar da Gaskiyar Duniya
Tabbacin fuska irin na FaceNet: wayoyi da ƙofofin fasfo suna tabbatar da ainihi ta hanyar duba idan fuskoki biyu sun faɗi tsakanin madaidaicin nesa.
Binciken samfur na gani: Shafukan kasuwancin e-commerce suna barin masu siyayya su loda hoto kuma su dawo da abubuwa masu kama da gani ta hanyar binciken makwabci mafi kusa.
Tabbatar da lasifika: mataimakan murya sun haɗa samfurin murya kuma su kwatanta shi da bayanan martaba don tabbatar da wanda ke magana.
Tabbatar da sa hannu da rubutun hannu: bankuna sun haɗa sa hannu da sa hannun tambaya da jarun tuta lokacin da tazarar ta wuce gefen koyo.
Hatsari & Tsare-tsare
Ƙungiyoyi daban-daban na iya amfani da kalmar iri ɗaya daban, don haka ayyana iyaka da wuri.
Alamomi na iya yin kama da ƙarfi yayin da aikin zahirin duniya bai yi daidai ba.
Yin watsi da ingancin bayanai da tsare-tsaren kimantawa galibi yana haifar da sakamako mara ƙarfi.
Taswirar Hanya
Fara da ma'anar harshe a sarari na sakamakon da kuke buƙata.
Zaɓi ma'aunin nasara ɗaya da yanayin gazawa ɗaya kafin gwaji.
Gudun ƙaramin matukin jirgi tare da bayanan wakilci, ba saitin demo da aka goge ba.
Takaddun inda Asarar Triplet da Ilimin awo ke taimakawa kuma inda mafi sauƙi hanyoyin suka fi kyau.
Ci gaba da Bincike
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 Triplet Loss and Metric Learning 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
Jagora na gaba
Siamese Networks da Asarar Sau Uku
Tambayoyin da ake yawan yi
What is Triplet Loss and Metric Learning?
Hasara sau uku tana koyar da hanyar sadarwa ta jijiyoyi don sanya abubuwa iri ɗaya kusa da juna da kuma abubuwa iri ɗaya nesa ba kusa ba a cikin sararin samaniya. Tushen da ke bayan gane fuska, binciken hoto, da tsarin shawarwari waɗanda ke buƙatar kwatanta abubuwa maimakon kawai rarraba su.
Wadanne bayanai guda uku ne suka hada da sau uku a cikin asarar sau uku?
Mai uku-uku ya ƙunshi anga, tabbataccen da ke raba ajin anga, da mara kyau daga wani aji daban.
Me yasa asarar sau uku ta haɗa da lokacin gefe?
Ba tare da tazara ba, ƙirar zata iya taswira komai zuwa wuri ɗaya, yana mai da duk nisa sifili kuma yana gamsar da hasara. Margin yana tilasta rabuwa ta gaske.
Wane sanannen tsarin 2015 ya shahara asara sau uku don gane fuska?
Google's FaceNet yayi amfani da hasarar sau uku don koyon ƙaramar fuska da saita ma'auni a fuskar tabbatarwa.
Menene ƙwararrun ƙirar ƙirar-ilimin awo don kowace shigarwa?
Samfurin yana tsara taswirorin abubuwan da za a shigar da vectors; sai ku kwatanta abubuwa ta hanyar auna tazara tsakanin abubuwan da aka saka su.
Me yasa ilimin awo ya dace da buɗaɗɗen matsaloli kamar ƙara sabbin fuskoki?
Saboda ganewa yana dogara ne akan nisa na sakawa, zaku iya yin rajistar sabon ainihi ta hanyar adana abin da aka saka ta kawai, ba a buƙatar sake horar da samfurin ba.