Code for paper: Gradient based sample selection for online continual learning Rahaf Aljundi, Min Lin, Baptiste Goujaud, Yoshua Bengio. Neurips 2019
- Python 2.8 or more.
- Pytorch 1.1.0
pip install -r requirements.txt
or
conda install --file requirements.txt
for GSS_IQP, please install MIXEDIP package: https://github.com/oxfordcontrol/miosqp following their README file. Note that for this you need to install quadprog and osqp
---- conda install -c omnia quadprog
---- conda install -c conda-forge/label/gcc7 osqp %% if it doesn't work check this https://anaconda.org/conda-forge/osqp
Demo:
For disjoint MNIST please run ./Disjoint_Mnist.sh
For permuted MNIST please run ./Permuted_Mnist.sh
For disjoint CIFAR-10 please run ./Cifar_10.sh