NoRD: A framework for noise-resilient self-distillation through relative supervision

Published at Applied intelligence Journal

A basic working code is given on CIFAR100 only. It's ablation study code. So, some of the losses are commented out. Link to the article and full code may be made available on request.

Installation

Mandatory Dependencies

• Python> 3.7
• PyTorch >= 1.7.0
• pandas
• scipy
• sklearn
• torchvision
• numpy

Usage

To run it

python teacher_n.py --arch ResNet34    --lr 0.01 --gpu-id 0
python student_n.py --t-path ./experiments/teacher_ResNet34_seed0/   --s-arch ResNet34    --lr 0.05 --gpu-id 0

Models can be downloaded from official Pytorch website or from SSKD Repository.

Contact

please contact saurabh_2021cs30@iitp.ac.in for any discrepancy.

Authors:

• Saurabh Sharma
• Shikhar Singh Lodhi
• Vanshika Srivastava
• Joydeep Chandra

Sources

SSKD Repository is used for the basic architecture.