Code of KDD2019 Paper Adaptive Deep Models for Incremental Learning: Considering Capacity Scalability and Sustainability
Official implementation for KDD 2019 paper Adaptive Deep Models for Incremental Learning: Considering Capacity Scalability and Sustainability.
Requirement
- python3.6.5
- Pytorch0.4.1
- numpy
- scipy
- sklearn
Files
--Incremental MNIST: Incremental MNIST dataset,including a training file and 4 testing files, which named as ldx_t.mat, lux_t.mat, rdx_t.mat and rux_t.mat.
--Draw.py: Draw the accuracy performance curve.
--Dataset.py: Prepare Incremental MNIST dataset.
--Test.py: Test performance of some model.
--IADM.py: A demo of our method.
Basic Usage
There are various parameters in the input structure paras:
--alp : Percentage of Fisher Information accumulated during Backpropagation.
--lr : Learning rate of our model
--drawstep: The frequence of Test duiring training (number of instance,default:12000)
--lamda : The regularization parameter in our paper.(In Eq(5).)
Quick Start
python IADM.py
Parameters/options can be tuned to get better results.
Citation
Please cite our work if you feel the paper or the code are helpful.
@inproceedings{yang2019adaptive,
author = {Yang, Yang and Zhou, Da-Wei and Zhan, De-Chuan and Xiong, Hui and Jiang, Yuan},
title = {Adaptive Deep Models for Incremental Learning: Considering Capacity Scalability and Sustainability},
booktitle = {KDD},
address = {Anchorage, AK},
pages = {74--82},
year = {2019}
}
Contact
If there are any questions, please feel free to contact with the authors: Da-Wei Zhou (zhoudw@lamda.nju.edu.cn) and Yang Yang (yangy@lamda.nju.edu.cn). Enjoy the code.