Samfuran Da Aka Gina Makamashi
Samfuran tushen makamashi (EBMs) suna koyon aikin 'makamashi' mai ƙima wanda ke ba da ƙarancin ƙima zuwa ingantaccen bayanai da ƙima masu girma zuwa bayanan da ba za a iya fahimta ba, yana bayyana yiwuwar rarraba ba tare da tilasta shi ya zama mai sauƙin daidaitawa ba.
Dubawa
This flexibility makes them a unifying lens for much of machine learning, from classifiers to generative models.
Zurfafa nutsewa
Samfurin tushen makamashi yana bayyana yuwuwar ta hanyar rarraba Boltzmann (Gibbs): p (x) daidai yake da exp (-E (x)), inda E (x) aikin makamashi ne da aka koya, galibi cibiyar sadarwa ne. Horowa yana tura kuzarin bayanan gaskiya kuma yana tura kuzarin komai. Abin kamawa shine aikin bangare Z, jimla ko haɗin kai na exp(-E(x)) akan duk abubuwan da za'a iya samu, wanda yawanci ba a iya ƙididdige shi. Don haka ana horar da EBMs tare da ƙima: bambance-bambancen banbance-banbance, daidaita maƙiya, ko ƙididdige surutu, kuma an ƙirƙira su ta hanyoyin MCMC kamar haɓakar Langevin wanda ke biye da ƙarfin kuzari. Misalai na yau da kullun sun haɗa da hanyoyin sadarwar Hopfield da Ƙuntataccen Injin Boltzmann; Ayyukan zamani suna haɗa EBMs zuwa nau'ikan watsawa, GANs, har ma da na'urori na yau da kullun waɗanda aka sake fassara su azaman ayyukan makamashi.
Fahimtar Fasaha
Samfurin yana ba da yuwuwar p(x) = exp(-E(x))/Z. Saboda Z (mai daidaitawa akan duk abubuwan da aka shigar) ba ya iya jurewa, ba kasafai kuke ƙididdige yiwuwar kai tsaye ba. Madadin haka, madaidaicin maki da samfurin Langevin yayi amfani da cewa gradient na log p(x) yayi daidai da -gradient na E(x), don haka Z ya fita. Langevin dynamics sannan yana haifar da samfurori ta hanyar nudging x sau da yawa cikin kuzari da ƙara hayaniya, tafiya zuwa yankuna masu ƙarancin ƙarfi, babban yuwuwar.
Dabarun Tasiri
Kudin da kasafin kuɗi
Hukunce-hukuncen gine-gine suna haifar da aiki da tsadar aiki na shekaru.
Shawarwari masu haske
Ilimin fasaha yana taimaka wa ƙungiyoyi su zaɓi tari mai kyau, ba kawai sabon abu ba.
Kula da inganci
Zaɓuɓɓukan injiniya mafi kyau suna rage abin dogaro a cikin samarwa.
Makomar Samfuran Tushen Makamashi
EBMs suna jin daɗin sabunta sha'awa saboda suna samar da gada mai ma'ana tsakanin samfuran watsawa, ƙirar ƙira mai ƙima, da cibiyoyin sadarwar wariya, ƙimar da samfurin watsawa ya koya shine ainihin ƙarfin kuzari. Yi tsammanin ƙarin tsarin matasan da ke amfani da ayyukan makamashi don sassauƙa, ƙayyadaddun ƙayyadaddun ƙayyadaddun ƙayyadaddun ƙayyadaddun ƙarfi (haɗa kuzari da yawa don tsara tsarawa), samfur mafi kyau da sauri fiye da MCMC, da aikace-aikace a cikin tunani da tsarawa inda 'nemo mafi ƙarancin ƙarfin kuzari' a zahiri yana nuna haɓakawa da ƙarancin gamsuwa.
Aiwatar da Gaskiyar Duniya
Cibiyoyin sadarwa na Hopfield suna aiki azaman ƙwaƙwalwar haɗin gwiwa waɗanda ke tuno tsarin da aka adana daga shigarwar hayaniya ko ɓangarori ta hanyar daidaitawa cikin yanayin ƙarancin kuzari.
Ƙuntataccen Injin Boltzmann da aka yi amfani da shi ta tarihi don tace haɗin gwiwa da kuma horar da cibiyoyin sadarwa mai zurfi.
Sake fassara ma'auni mai rarrabawa azaman ƙirar tushen makamashi (tsarin JEM) don haɓaka daidaitawa, ƙarfi, da ganowa ba-rarraba
Hasashen da aka tsara da gamsuwa, inda ake samun mafita ta hanyar rage ƙarfin da aka koya akan yawancin masu mu'amala (misali, ƙima ko shimfidawa)
Hatsari & Tsare-tsare
Haɓaka ma'auni ɗaya na iya ɓoye manyan raunin tsarin.
Sau da yawa ana raina kayan more rayuwa da kuma kuɗin kulawa.
Tsaro da gibin lura na iya girma yayin da tsarin ke ƙara haɓaka.
Taswirar Hanya
Ƙayyade latency, inganci, da maƙasudin farashi kafin aiwatarwa.
Alamar ma'auni a ƙarƙashin ainihin kaya da yanayin bayanai.
Kula da kayan aiki don kurakurai, ɗigo, da tasirin mai amfani.
Shirya bijirowa da hanyoyin mayar da martani kafin sikeli.
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 Energy-Based Models 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
Samfuran Ƙarfafa Maki
Tambayoyin da ake yawan yi
What is Energy-Based Models?
Samfuran tushen makamashi (EBMs) suna koyon aikin 'makamashi' mai ƙima wanda ke ba da ƙarancin ƙima zuwa ingantaccen bayanai da ƙima masu girma zuwa bayanan da ba za a iya fahimta ba, yana bayyana yiwuwar rarraba ba tare da tilasta shi ya zama mai sauƙin daidaitawa ba. Wannan sassauci yana sa su zama ruwan tabarau mai haɗa kai don yawancin koyo na inji, daga masu ƙira zuwa ƙirar ƙira.
A cikin samfurin tushen makamashi, ta yaya makamashi ke da alaƙa da yuwuwar wurin bayanai?
Ta hanyar rarraba Boltzmann, p (x) yayi daidai da exp(-E (x)), don haka ana sanya bayanai masu inganci da ƙarancin ƙarfi da yuwuwar girma.
Menene ya sa horar da ƙirar makamashi mai wahala?
Ƙididdigar Z na buƙatar tarawa ko haɗawa da exp(-E(x)) a kan dukkan sararin shigar da bayanai, wanda gabaɗaya ba zai yuwu ba, tilasta kusan hanyoyin horo.
Wace hanyar samfur ake amfani da ita don zana samfurori daga EBM?
Halin Langevin yana motsa samfurori zuwa ƙasa tare da ƙarar kuzari yayin ƙara hayaniya, yana haɗuwa zuwa yankuna masu ƙarancin ƙarfi.
Me yasa hanyoyin kamar daidaita maki zasu iya guje wa ƙididdige aikin ɓangaren Z?
Tun da Z bai dogara da x ba, banbance log p(x) dangane da x ya soke shi, yana barin ƙarar makamashi kawai, wanda za'a iya gani.
Wanne daga cikin waɗannan shine ƙirar tushen makamashi na gargajiya?
Cibiyoyin sadarwa na Hopfield wani samfurin tushen makamashi ne na farko wanda ke adana alamu azaman jihohi masu ƙarancin kuzari kuma suna tunawa da su ta hanyar rage ƙarfi.