Robust Few-shot Learning Without Using any Adversarial Samples - Official Implementation
(Accepted to be published in T-NNLS)

Dependencies:

1. torch 1.10.2
2. torchvision 0.11.3
3. torchattacks 3.2.4
4. tqdm 4.63.0
5. fastai 1.0.58

---

Dataset Preparation

1. Download the dataset from here and unzip it in ./CIFAR-FS/
2. Prepare the dataset using ./CIFAR-FS/prepare.py

---

Running Experiments

Step-1. Pretraining Stage

1. Train the teacher model using ./scripts/pretrain_teacher.sh
2. Train the student model using ./scripts/pretrain_student.sh

Step-2. Finetuning & Evaluation Stage

1. Finetune and evaluate the pretrained student model using ./scripts/finetune.sh

---

Acknowledgements

This repo is adapted from Dhillon et al. 2020 and Wang et al. 2020