Awesome
Condensed Composite Memory Continual Learning
This repository contains all the code used for the creation of the paper "Condensed Composite Memory Continual Learning" published at IJCNN 2021 (https://ieeexplore.ieee.org/document/9533491).
Requirements
In order to install all required packages automatically run pip install -r requirements.txt.
Running the code
Each experiment for a certain method and dataset can be run using the corresponding python script. Results will be saved in the logs folder.
Experience Replay
- 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
BiC
- 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
Dataset Condensation
- 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
Condensed Composite Memory Continual Learning
- 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