Awesome

Online Class-Incremental Continual Learning with Adversarial Shapley Value

Notes !!

• This repository contains the TensorFlow implementation of ASER and other baselines. The results in the paper can be reproduced by following the instructions below.
• PyTorch implementation of ASER and more baselines can be found in this repository. Note that the PyTorch version is more efficient than the original TensorFlow implementation and has better performance.

Requirements

conda env create -f environment.yml

Data

• CIFAR10 & CIFAR100 will be downloaded during the first run
• Mini-ImageNet: Download from https://www.kaggle.com/whitemoon/miniimagenet/download , and place in Data/miniimagenet/

Running Experiments

• ASER = Adversarial Shapley Value Experience Replay
• AGEM = Averaged Gradient Episodic Memory
• ER = Experience Replay
• EWC = Elastic Weight Consolidation
• MIR = Maximally Interfered Retrieval
• GSS = Gradient-Based Sample Selection

To reproduce the result in the paper:

source reproduce.sh