Ci bi mu melGis bi ci topp
Etiketu ëmbiitu IA ci platform sosio yi
Askan wi
GUIDE bu am solo
Nooyal etiket
Label smoothing replaces hard one-hot targets with a softened target distribution during training.
Déglul tegtal biiIA-nettali · ~2 min
Ci xët wii2 simili jàng
Résumé
It can regularize some models, but its effects on accuracy, calibration, and distillation depend on the task, model, and evaluation.
Plongeur bu xóot
With K classes, one common label-smoothing convention assigns 1 − ε to the correct class and distributes ε across the remaining classes; implementations may instead distribute ε across all classes, so check the loss definition. This prevents the target from demanding probability one for the labeled class and can discourage extreme logits. The Inception-v3 work and the Transformer paper are prominent documented uses of the method. Research has found calibration and generalization benefits in some settings, while also showing that smoothing can remove information about similarities among incorrect classes and weaken a model as a knowledge-distillation teacher. It is therefore not a universal accuracy or calibration improvement. Choose ε and evaluate calibration, task quality, and downstream distillation on representative validation data.
Doxal ci àdduna dëgg
Xeetu ImageNet: Inception-v3 dafa jëfandikoo etiketu nooy (epsilon 0.1) ngir gëna dëgëral top-1 ak wàññi wóolu sa bopp bu ëpp.
Tekki masin: Transformer bi njëkk dafa def etiketu nooy bu 0.1, di jaay tuuti jaaxle ngir am poñ yu gëna rëy ci BLEU.
Xamme kàddu yi: target yu nooy yi dañuy wàññi njuumte yu gëm seen bopp ba noppi gëna suqali etalonnage ci audio bu bari bruit bi.
Modèlu nataalu pajum: nooyloo dafay joxe probabilite yu gëna kalibre, lu am solo sudee poñ wóolu sa bopp mooy leeral dogal yi ci pajum.
Ëlëgu nooy etiketu
Nooyal etiketu mingi wéy di nekk lu ñuy faral di def ci tàggat-yaram bu mag, waaye gëstu baa ngi jëm ci nooyloo buy méngoo ak jàng buy méngale epsilon ci misaal bu nekk wala ci klaas bu nekk, du jëfandikoo benn valeur bu plat. Pexe yu lalu ci etalonnage yu melni perte focal ak scaling température dañu leen di faral di méngale wala ñu boole ci. Lu model yi di màgg, te xayma yu wóorul yi dañuy gëna am solo ci wàllu kaaraange, xaaral nooyloo nekk benn jumtukaay ci lu bari ngir génne poñ yu wóor, ak bàyyi xel bu baax ci jafe-jafe yi mu xamee ak distilaasioŋ.
Teg ko ci jëf
Lu tax mu am solo
- dogal yu gëna leer. Daf lay jàppale nga tàqale kàddu yu leer ci wàllu xarala ak làkku fësal njaay.
- Njëgg ak budget. Mën nga laaj laaj yu gëna baax ci samp gi balaa ngay dugal xaalis wala sa jotu liggéey.
- Ekip ak def liggéey. Ekip yi bokk xam-xam ñoo gëna mëna jël yenn dogal ci wàllu produit, politik ak jàng.
Moytul
- Ekip yu bari mën nañu jëfandikoo benn baat ci anam wu wuute, kon teela leeral yaatuwaayam.
- Benchmark yi mën nañu nuru lu am doole waaye performance yi ci àdduna bi duñu tolloo.
- Bëgg kalite done ak palaŋu jàngat dafay faral di jur njariñ yu yomba dagg.
Jéego yi ci topp
- Tàmbaleel ci joxe leeral ci làkk wu leer ci njariñ li nga soxla.
- Tannal benn metric bu baax ak benn anam bu baaxul balaa ngay saytu.
- Doxal ab pilote bu ndaw ak ay done yu representatif, du ab demo bu leer.
- Bindal fi Label Smoothing di jàppale ak fi pexe yu gëna yomba gëna baax.
Weyal di banneexu
Bulletin bu amul fayda
Jàngal xibaar AI bu bes
Ñatti jaar-jaari IA yuñ firndeel bis bu nekk ci ayu-bis bi, ñu bind leen ci Àngle bu leer. Gratis ba fàww, amul yëgle.
Benn email bis bu nekk ci ayu-bis bi. Benn klik nga dindi abonemaa bi. Dunu musa jaay wala séddoo sa adres.
Nattal sa bopp
Take the Label Smoothing quiz
Feedback ci saasi ci bépp tontu, ak sertifikaa buñ mëna séddoo ak ID buñ mëna saytu soo jàllee ci njàng mi.
Jàppale njàngum IA te doo fay. AI Understanding mbootaay la buy def te Yàlla tax — amul yëgle, amul paywall, musul am. Defal ab sarax
Laaj yi ñuy faral di laaj
Luy nooy etiket?
Label smoothing replaces hard one-hot targets with a softened target distribution during training. It can regularize some models, but its effects on accuracy, calibration, and distillation depend on the task, model, and evaluation.
Lan la etiketu nooy soppi ci wàllu tàggat yaram?
Lu moy 1.0/0.0 bu dëgër, etiketu nooy dafay jox (1 - epsilon) ci klaas bu dëggu bi ba noppi tasaare epsilon ci ñeneen ñi.
What is one intended effect of replacing hard one-hot targets with smoothed targets?
Smoothing softens the target distribution so the training objective does not demand a one-hot probability vector.
Which landmark architectures are prominent examples of label-smoothing use?
Inception-v3 and the Transformer paper are prominent documented uses of label smoothing; this wording makes no claim that either originated the technique.
What can label smoothing do to calibration?
Research reports calibration effects that depend on the model and evaluation setting; label smoothing is not a universal calibration guarantee.
Lan mooy njuumte biñ bind ci nooy etiketu?
Suñu gënee dëgëral clusters yi ak dakkal diggante yu yomb yi am ci diggante klaas yi, nooy mën na dindi xibaar yi distillation xam-xam bi di aju ci.
Weyal di jàng
Gid yu jëm ci loolu
Tann nañu yeneen njiit ngir topic bii