cliffordlai/apd

"Scalable and Order-robust Continual Learning with Additive Parameter Decomposition", ICLR 2020

"Scalable and Order-robust Continual Learning with Additive Parameter Decomposition", ICLR 2020

• Jaehong Yoon(KAIST), Saehoon Kim(AITRICS), Eunho Yang(KAIST, AITRICS), and Sung Ju Hwang(KAIST, AITRICS)

This project hosts the code for our ICLR 2020 paper.

While recent continual learning methods largely alleviate the catastrophic problem on toy-sized datasets, some issues remain to be tackled to apply them to real-world problem domains. First, a continual learning model should effectively handle catastrophic forgetting and be efficient to train even with a large number of tasks. Secondly, it needs to tackle the problem of order-sensitivity, where the performance of the tasks largely varies based on the order of the task arrival sequence, as it may cause serious problems where fairness plays a critical role (e.g. medical diagnosis). To tackle these practical challenges, we propose a novel continual learning method that is scalable as well as order-robust, which instead of learning a completely shared set of weights, represents the parameters for each task as a sum of task-shared and sparse task-adaptive parameters. With our Additive Parameter Decomposition (APD), the task-adaptive parameters for earlier tasks remain mostly unaffected, where we update them only to reflect the changes made to the task-shared parameters. This decomposition of parameters effectively prevents catastrophic forgetting and order-sensitivity, while being computation- and memory-efficient. Further, we can achieve even better scalability with APD using hierarchical knowledge consolidation, which clusters the task-adaptive parameters to obtain hierarchically shared parameters. We validate our network with APD, APD-Net, on multiple benchmark datasets against state-of-the-art continual learning methods, which it largely outperforms in accuracy, scalability, and order-robustness.

Reference

If you use this code as part of any published research, please refer the following paper.

@inproceedings{yoon2020apd,
  title={Scalable and Order-robust Continual Learning with Additive Parameter Decomposition},
  author={Yoon, Jaehong and Kim, Saehoon and Yang, Eunho and Hwang, Sung Ju},
  year={2020},
  booktitle={ICLR}
}

Prerequisites

• Python 3.x
• Tensorflow 1.14.0

Dataset Preparation

We give example codes for CIFAR-100-Split / CIFAR-100-Superclass dataset. Download CIFAR-100-python version dataset (Toronto-CIFAR).

Model Run

• Run one of the followings.
• Default task order is the "orderA". you can modify "order_type" in the code.

CIFAR-100-Split experiment

# 10 classes & 10 tasks, APD(2) which includes hierarchical knowledge consolidation
$ python cifar100_apd_bash.py

CIFAR-100-Superclass experiment

# 5 classes & 20 tasks, APD(2) which includes hierarchical knowledge consolidation
$ python cifar100_sup_apd_bash.py

Authors

Jaehong Yoon¹, Saehoon Kim², Eunho Yang¹², and Sung Ju Hwang¹²

¹KAIST @ School of Computing, KAIST, Daejeon, South Korea

²AITRICS @ Seoul, South Korea