Domain Adaption in One-Shot Learning

Implementation of Domain Adaption in One-Shot Learning.

Paper

You can find our paper at Springer or PDF.

Citation

If you find DAOSL useful in your research, please consider to cite:

@inproceedings{dong2018domain, title={Domain adaption in one-shot learning}, author={Dong, Nanqing and Eric P. Xing}, booktitle={Joint European Conference on Machine Learning and Knowledge Discovery in Databases}, pages={573--588}, year={2018}, organization={Springer} }

Setup

Requirements

python 3.5
tensorflow 1.8
CUDA  9.0
cuDNN 7.0

Installation

sh setup.sh

Training

Train one-shot classifier for 5-way 1-shot learning.

python3 convert_data.py --data-name=omniglot
python3 convert_data.py --data-name=emnist --num-target-examples=20
python3 train_one_shot.py --exp-name=one_shot --source=omniglot --target=emnist_20 --num-ways=5

Train adversarial domain adaption (ADA) for 5-way 1-shot learning.

python3 convert_data.py --data-name=omniglot
python3 convert_data.py --data-name=emnist --num-target-examples=20
python3 train_ada.py --exp-name=ada --source=omniglot --target=emnist_20 --num-ways=5 --la=0.001

Train adversarial domain adaption (ADA) with reinforced sample selection (RSS) for 5-way 1-shot learning.

python3 convert_data.py --data-name=chars
python3 convert_data.py --data-name=sim
python3 convert_data.py --data-name=dis
python3 train_rss.py --exp-name=rss --num-ways=5 --la=0.001 --gamma=0.1