This repository contains all the code used for the creation of the paper "Condensed Composite Memory Continual Learning" accepted for publication at IJCNN 2021.
In order to install all required packages automatically run pip install -r requirements.txt.
Each experiment for a certain method and dataset can be run using the corresponding python script. Results will be saved in the logs folder.
- MNIST dataset:
python run_experience_replay_MNIST.py - FashionMNIST dataset:
python run_experience_replay_FashionMNIST.py - SVHN dataset:
python run_experience_replay_SVHN.py - CIFAR10 dataset:
python run_experience_replay_CIFAR10.py
- MNIST dataset:
python run_BiC_MNIST.py - FashionMNIST dataset:
python run_BiC_FashionMNIST.py - SVHN dataset:
python run_BiC_SVHN.py - CIFAR10 dataset:
python run_BiC_CIFAR10.py
- MNIST dataset:
python run_compressed_buffer_MNIST.py - FashionMNIST dataset:
python run_compressed_buffer_FashionMNIST.py - SVHN dataset:
python run_compressed_buffer_SVHN.py - CIFAR10 dataset:
python run_compressed_buffer_CIFAR10.py
- MNIST dataset:
python run_compositional_buffer_MNIST.py - FashionMNIST dataset:
python run_compositional_buffer_FashionMNIST.py - SVHN dataset:
python run_compositional_buffer_SVHN.py - CIFAR10 dataset:
python run_compositional_buffer_CIFAR10.py