This repository was a part of the final hand-in for course no. 02456 - Deep Learning at the Technical University of Denmark.
We attempt to recreate the results of Lu et al 2022 in which they have created and implemented the iCaRL algorithm to gurantee out-of-distribution generalization.
In order to run the jupyter notebooks you must install the following packages
- Seaborn
- Matplotlib
- Scikit-image
- Scikit-learn
- Pytorch
- Numpy
These packages can be installed either via git | pip.
We recommend setting up a virtual enviroment - Read more here
The classification of CMNIST is computationally heavy - thus we recommend training with a GPU. The requirements installer install pytorch cuda.
For automaticly installing the required packages please enter the following in a terminal:
$ pip install -r requirements.txt
The PC algorithm works from R - Requirering additional packages. More can be found here
we recommend installing BioCManager to install plugins for R to work with the pcalg package.
- Morten Johnsen (s184334)
- Peter Emil Carstensen (s184332)
- Jacob Bendsen (s184328)
- Marie Pedersen (s174329)
- Frederik Larsen (s174159)
To reproduce our results please run run the following files inside the code folder:
- CMNIST_main.ipynb
- Synthetic_main.ipynb
Please run dataloader.py