search

yuyinzhou / L2B

This repository includes the official project of L2B, from our paper "Learning to Bootstrap for Combating Label Noise".
28 stars 3 forks source link

Hi Zhou, i want to know wheather the asymmetric label noise is generated like previous method UNICON? #4

Closed LanXiaoPang613 closed 2 weeks ago

LanXiaoPang613 commented 2 weeks ago

Hi Zhou, i want to know wheather the asymmetric label noise is generated like previous method unicon?Actually, I saw flip and flip2 in the code, but I did not see the asymmetric label noise generation function used in unicon, such as cat<->dog, bird->plane, etc.

LanXiaoPang613 commented 2 weeks ago

besides, when i read the code, i found that the loss l_f_meta is zero when multiplied by eps. It indicates that the gradient is 0. So i want to understand the purpose of this operation. image

yuyinzhou commented 2 weeks ago

Thanks for your interest in our work!

  1. We follow previous implementation in dividemix to generate asymmetric noise.

  2. For the initialization of l_f_meta, we follow the approach of Ren et al. in "Learning to Reweight Examples for Robust Deep Learning," where the initialization value is set to zero.