Label Noise

تشويش التسميات

افتراض أن تسميات الإنسان تحتوي على نسبة ثابتة من الأخطاء العشوائية (10% في هذه الورقة). يُدمج في دالة الفقد لمنع النموذج من الثقة المفرطة في مقارنات فردية.

An assumption that human labels contain a fixed rate of random errors (10% in this paper). Incorporated into the loss function to prevent the model from over-trusting individual comparisons.

Also translated asضجيج التسميات

First appears in this corpus in: A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting (1997)

Appears in these papers

A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting1997 in the sky ✦
Sharpness-Aware Minimization for Efficiently Improving Generalization2021 in the sky ✦
Sharpness-Aware Minimization for Efficiently Improving Generalization2021 in the sky ✦