maxgrad

This repository contains the source code accompanying our CVPR 2021 paper.

Gradient-Based Algorithms for Machine Teaching
Pei Wang, Kabir Nagrecha, Nuno Vasconcelos.
In CVPR, 2021.

@InProceedings{wang2021gradient,
author = {Wang, Pei and Nagrecha, Kabir and Vasconcelos, Nuno},
title = {Gradient-Based Algorithms for Machine Teaching},
booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2021}
}

Requirements

1. The project was implemented and tested in Python 3.5 and Pytorch 1.0. Other versions should work after minor modification.
2. NVIDIA GPU and cuDNN are required to have fast speeds. For now, CUDA 8.0 with cuDNN 6.0.20 has been tested. The other versions should be working.

Datasets

Butterflies and Chinese Characters are used. Please organize them as below after download,

datasets
|_ butterflies_crop
  |_ images
    |_ Viceroy
    |_ ...

datasets
|_ chinese_chars
  |_ images
    |_ grass
    |_ ...

Implementation details

To reproduce results of our method on simulated learners

train_butterflies_maxgrad.py
train_chineseChars_maxgrad.py

Our selected teaching images are contained in

butterflies_Lt_gt_tr.txt
ChineseChars_Lt_gt_tr.txt

Time and Space

All experiments were run on NVIDIA TITAN Xp

1. butterflies

model	#GPUs	train time
train_butterflies_maxgrad	1	~5min

2. Chinese characters

model	#GPUs	train time
train_chineseChars_maxgrad	1	~3min

Teaching system

system

Disclaimer

For questions, feel free to reach out

Pei Wang: [email protected]