This code allows to study active learning methods on several convolutional neural networks, with diferent image datasets, and other parameters, at once.
At the end of the experiment it shows the results in one or several plots depending on user arguments, and save them.
Therefore you can play with the results after, without relaunching all the experiment, that can be pretty long.
Active learning methods: least confidence sampling, margin sampling, and entropy sampling.
Convolutional Neural Networks: VggNet, ResNet and AlexNet.
Image datasets: CIFAR10 and CIFAR100.
Open a terminal, and:
git clone https://github.com/rafutek/CNN-active-learning.gitcd CNN-active-learningpip install -r requirements.txtcd codepython experiment.py -hto display the help
Read the wiki for more informations
Special thanks to our professor Pierre-Marc Jodoin for its teaching, without him this project would not exist.