Imọ Itọsọna
Barlow Twins ati Apọju Idinku
Barlow Twins jẹ ọna abojuto ti ara ẹni ti o kọ ẹkọ awọn aṣoju nipa ṣiṣe matrix ibamu-agbelebu laarin awọn iwo afikun meji ti o sunmọ si matrix idanimọ.
Gbọ si itọsọna yiiAI-ṣe afihan · ~2 min
Lori iwe yi2 min ka
Akopọ
O yago fun iṣubu nipasẹ ipilẹ-idinku-idinku dipo awọn odi tabi awọn koodu ipanu.
Jin Dive
Dabaa nipasẹ Facebook AI ni 2021 ati oniwa lẹhin neuroscientist H. Barlow ká apọju-idinku opo, Barlow Twins ifunni meji daru wiwo ti ohun image nipasẹ awọn nẹtiwọki kanna lati gbe awọn ipele meji ti awọn ifibọ. O ṣe iṣiro matrix ibamu-agbelebu laarin awọn paati ti awọn ifisinu ifisinu meji wọnyi, ni iwọn kọja ipele naa. Idi ti nfa matrix yii si ọna idanimọ: awọn titẹ sii diagonal yẹ ki o jẹ 1 (ẹya-ara kọọkan jẹ aibikita si augmentation) ati awọn titẹ sii ita-rọsẹ yẹ ki o jẹ 0 (awọn ẹya ara ẹrọ ti o yatọ si jẹ ohun ọṣọ, idinku apọju). Oro on-rọsẹ fi agbara mu aiṣedeede; Oro idinku-idinku ni pipa-rọsẹ-rọsẹ nipa ti ṣe idilọwọ iṣubu nitori awọn ẹya ti ohun ọṣọ ko le jẹ aami kanna. Ko dabi BYOL ko nilo asymmetry, asọtẹlẹ, tabi idaduro-gradient, ati pe ko dabi SimCLR ko nilo awọn orisii odi, botilẹjẹpe o ni anfani lati awọn ifibọ iwọn-giga.
Real-World imuse
Awọn koodu koodu iṣaju ikẹkọ ti o mu awọn ẹya ti o ni ibatan si awọn ẹya ti o wulo fun isọdi isalẹ pẹlu data aami to lopin.
Ikẹkọ lori ohun elo iwọntunwọnsi nibiti awọn ipele odi nla ko wulo, nitori Barlow Twins jẹ aibikita iwọn-iwọn.
Ṣiṣẹda iwapọ, awọn ifibọ aisi-padabọ fun ikojọpọ tabi wiwa anomaly ni aworan sensọ ile-iṣẹ.
Ṣiṣẹ bi ipilẹ-abojuto ti ara ẹni ninu iwadi ti o ṣe afiwe awọn ilana imukuro-idasilẹ kọja SimCLR, BYOL, ati VICReg.
Ojo iwaju ti Barlow Twins ati Idinku Apọju
Barlow Twins ṣe iranlọwọ lati tan idile alaye-imọran ti awọn ọna abojuto ti ara ẹni, pataki julọ VICReg, eyiti o ya iyatọ, aibikita, ati awọn ofin ibaramu ni gbangba. Reti idinku-idinku ati awọn ibi-afẹde ẹya-ara lati tọju ifitonileti bii a ṣe kọ awọn koodu koodu ti o ṣe agbejade iwapọ, awọn ẹya ti kii ṣe laiṣe, ati lati fa kọja awọn aworan sinu multimodal ati awọn eto jara-akoko nibiti awọn ohun ọṣọ, awọn aṣoju to lagbara ṣe iranlọwọ fun awọn awoṣe isalẹ lati kọ ẹkọ lati awọn aami to lopin.
Fifi o sinu iwa
Kini idi ti o ṣe pataki
- Iye owo ati isuna. Awọn ipinnu faaji ṣe awakọ iṣẹ ati idiyele iṣẹ fun awọn ọdun.
- Awọn ipinnu diẹ sii. Ẹkọ imọ-ẹrọ ṣe iranlọwọ fun awọn ẹgbẹ lati yan akopọ to tọ, kii ṣe ọkan tuntun nikan.
- Iṣakoso didara. Awọn yiyan imọ-ẹrọ to dara julọ dinku awọn iṣẹlẹ igbẹkẹle ni iṣelọpọ.
Ṣọra fun
- Ṣiṣepe ala-ilẹ kan le tọju awọn ailagbara eto ti o gbooro.
- Awọn ohun elo amayederun ati awọn idiyele itọju nigbagbogbo ni aibikita.
- Aabo ati awọn ela akiyesi le dagba bi awọn eto ṣe di eka sii.
Next awọn igbesẹ
- Ṣetumo lairi, didara, ati awọn ibi-afẹde idiyele ṣaaju imuse.
- Aṣepari labẹ ẹru ojulowo ati awọn ipo data.
- Abojuto ohun elo fun awọn aṣiṣe, fiseete, ati ipa olumulo.
- Mura ipadasẹhin pada ati awọn ipa ọna esi iṣẹlẹ ṣaaju iwọn.
Tesiwaju Ṣiṣawari
Ìwé ìròyìn ọfẹ
Gba ìkànìyàn AI ojoojúmọ́
Ìtàn AI mẹta tó jẹ́rìí ní gbogbo owurọ́ ọjọ́ iṣẹ́, kọ́wé ní Gẹ̀ẹ́sì tó rọrùn. Ọfẹ́ fún gbogbo àkókò, kò sí ìpolówó.
Ìmeèlì kan ní gbogbo ọjọ́ iṣẹ́. Yọkúrò ní ìkànìyàn ní ìkànìyàn kan. A kò ta tàbí pín àdírẹ́sì rẹ.
Ṣe idanwo fun ara rẹ
Take the Barlow Twins and Redundancy Reduction quiz
Ìdáhùn lẹ́sẹ̀kẹsẹ̀ lórí gbogbo ìdáhùn, àti iwe-ẹ̀rí tó lè pin pẹ̀lú ID tó lè jẹ́rìí lẹ́yìn tí o bá kọ́ ẹ̀kọ́.
Ṣe àtìlẹ́yìn ẹ̀kọ́ AI ọfẹ́. AI Understanding jẹ́ àjọ àìní èrè 501(c)(3) — kò sí ìpolówó, kò sí àkúnya owó, ní gbogbo àkókò. Ṣe ẹbun kan
Awọn ibeere ti a beere nigbagbogbo
Kini Barlow Twins ati Idinku Apọju?
Barlow Twins jẹ ọna abojuto ti ara ẹni ti o kọ ẹkọ awọn aṣoju nipa ṣiṣe matrix ibamu-agbelebu laarin awọn iwo afikun meji ti o sunmọ si matrix idanimọ. O yago fun iṣubu nipasẹ ipilẹ-idinku idinku dipo awọn odi tabi awọn koodu ipanu.
What does Barlow Twins compare during representation learning?
The method forms a cross-correlation matrix from representations of two distorted views.
What target does the Barlow Twins objective push the cross-correlation matrix toward?
The paper defines the objective by making the cross-correlation matrix close to identity.
What does a diagonal term in the objective encourage?
The diagonal term encourages invariance between matching feature components.
What does the off-diagonal penalty encourage?
The method reduces redundancy by pushing off-diagonal correlations toward zero.
How does Barlow Twins differ from methods that require negative pairs?
The original Barlow Twins paper describes avoiding negatives and matching the cross-correlation matrix to identity.
Tesiwaju kikọ
Jẹmọ awọn itọsọna
Awọn itọsọna diẹ sii ti a yan fun koko yii