Up tókànItọsọna atẹle
The Explainability vs Accuracy Tradeoff
Awọn ipilẹ
Awọn ipilẹ Itọsọna
Iyatọ-Itọpa Iṣowo
Iṣowo-iyatọ-iyatọ n ṣalaye idi ti awoṣe le kuna nipa jijẹ rọrun pupọ tabi eka pupọ.
Gbọ si itọsọna yiiAI-ṣe afihan · ~2 min
Lori iwe yi2 min ka
Akopọ
It's the central tension behind underfitting versus overfitting, and getting it right determines whether your model generalizes to new data.
Jin Dive
Gbogbo aṣiṣe asọtẹlẹ ti awoṣe ṣe le pin si awọn ẹya mẹta: irẹjẹ, iyatọ, ati ariwo ti ko le dinku. Irẹjẹ jẹ aṣiṣe lati awọn arosinu ti ko tọ - awoṣe ti o rọrun pupọ lati mu ilana gidi, bii ibamu laini taara si ti tẹ (aiṣedeede). Iyatọ jẹ aṣiṣe lati ifamọ si apẹẹrẹ ikẹkọ kan pato - awoṣe ti o rọ ti o ṣe akori awọn quirks ati ariwo (overfitting). Awọn apeja ni wipe sokale ọkan duro lati gbe awọn miiran. Oniyepo ilopo giga-giga npa ojuṣaaju ṣugbọn awọn asọtẹlẹ rẹ n yipada lainidi pẹlu ipilẹ data tuntun kọọkan. Ibi-afẹde kii ṣe lati yọkuro boya aṣiṣe ṣugbọn lati wa aaye didùn nibiti apao wọn - lapapọ aṣiṣe ti a nireti lori data ti a ko rii - kere julọ.
Real-World imuse
Yiyan ijinle ti igi ipinnu: igi aijinile ti o wa labẹ (irẹjẹ giga), igi ti o jinlẹ pupọ ṣe iranti awọn ori ila ikẹkọ (iyatọ giga), nitorina o tunse ijinle nipasẹ aṣiṣe afọwọsi.
Ṣiṣeto agbara isọdọtun (lambda) ni oke tabi isọdọtun lasso lati ṣe iṣowo ilosoke kekere ni irẹjẹ fun idinku nla ninu iyatọ ati deede idanwo to dara julọ.
Lilo awọn igbo laileto, eyiti o jẹ aropin ọpọlọpọ awọn igi iyatọ-giga ti o ni ibatan lati dinku iyatọ gbogbogbo laisi infating abosi pupọ.
Yiyan nọmba awọn aladugbo k ni k-NN: k = 1 ni iyatọ ti o ga ati ti o tẹle ariwo, lakoko ti k ti o tobi pupọ bori ati ṣe afikun abosi.
Ojo iwaju ti Ijaja-Iyatọ Iṣowo
Ẹkọ ti o jinlẹ ti ṣe idiju itan-akọọlẹ Ayebaye. Awọn oniwadi ṣe akiyesi “isọkalẹ ilọpo meji,” nibiti aṣiṣe idanwo ti kọkọ dide, lẹhinna ṣubu lẹẹkansi bi awọn nẹtiwọọki ti o pọ ju ti o pọ ju ti o ti kọja ẹnu-ọna interpolation - o dabi ẹni pe o tako U-curve. Loye idi ti awọn awoṣe nla ṣe gbogbogbo laibikita aṣiṣe ikẹkọ odo-odo jẹ aala iwadii ti nṣiṣe lọwọ, ti a so si isọdọtun alaiṣedeede lati awọn iṣapeye bii SGD. Awọn oṣiṣẹ adaṣe n gbilẹ si titọkasi agbara, awọn ofin iwọn, ati awọn ifọwọyi afọwọsi dipo iṣowo iwe kika nikan.
Fifi o sinu iwa
Kini idi ti o ṣe pataki
- Awọn ipinnu diẹ sii. O ṣe iranlọwọ fun ọ lati ya sọtọ awọn iṣeduro imọ-ẹrọ lati ede tita.
- Iye owo ati isuna. O le beere awọn ibeere imuse to dara julọ ṣaaju lilo owo tabi akoko.
- Ẹgbẹ ati ṣiṣan iṣẹ. Awọn ẹgbẹ pẹlu oye pinpin ṣe ọja to dara julọ, eto imulo, ati awọn ipinnu ikẹkọ.
Ṣọra fun
- Awọn ẹgbẹ oriṣiriṣi le lo ọrọ kanna ni oriṣiriṣi, nitorinaa ṣalaye iwọn ni kutukutu.
- Awọn aṣepari le wo lagbara lakoko ti iṣẹ-aye gidi ko ṣe deede.
- Aibikita didara data ati awọn ero igbelewọn nigbagbogbo ṣẹda awọn abajade ẹlẹgẹ.
Next awọn igbesẹ
- Bẹrẹ pẹlu itumọ-ede itele ti abajade ti o nilo.
- Mu metiriki aṣeyọri kan ati ipo ikuna kan ṣaaju idanwo.
- Ṣiṣe awakọ kekere kan pẹlu data aṣoju, kii ṣe eto demo didan.
- Iwe ibi ti Bias-Variance Tradeoff ṣe iranlọwọ ati nibiti awọn ọna ti o rọrun dara julọ.
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 Bias-Variance Tradeoff 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
What is Bias-Variance Tradeoff?
Iṣowo-iyatọ-iyatọ n ṣalaye idi ti awoṣe le kuna nipa jijẹ rọrun pupọ tabi eka pupọ. O jẹ ẹdọfu aringbungbun lẹhin isọdi labẹ fifin, ati gbigba ni ẹtọ pinnu boya awoṣe rẹ ṣe akopọ si data tuntun.
Oju iṣẹlẹ wo ni ami Ayebaye ti iyatọ giga (overfitting)?
Iyatọ ti o ga julọ fihan bi aafo nla laarin iṣẹ ikẹkọ ti o dara julọ ati iṣẹ idanwo ti ko dara - awoṣe ti ṣe iranti data ikẹkọ.
In the classic bias-variance picture described in the guide, which tradeoff can accompany increased model flexibility?
In the classic account, greater flexibility can reduce bias while increasing sensitivity to the sample. The guide also describes double descent, so this is not a universal monotonic law.
In a squared-error decomposition, squared bias is 2, variance is 3, and irreducible noise is 1. What is the total expected error?
The supplied bias term is already squared. Add the three components: 2 + 3 + 1 = 6.
Why can averaging many de-correlated trees help in the random-forest example?
The guide describes averaging high-variance, de-correlated trees as a way to reduce ensemble variance.
Tesiwaju kikọ
Jẹmọ awọn itọsọna
Awọn itọsọna diẹ sii ti a yan fun koko yii