LAMDA: Label Matching Deep Domain Adaptation

This is the implementation of the paper LAMDA: Label Matching Deep Domain Adaptation which has been accepted at ICML 2021.

A. Setup

A.1. Install Package Dependencies

Install manually

Python Environment: >= 3.5
Tensorflow: >= 1.9

Install automatically from YAML file

pip install --upgrade pip
conda env create --file tf1.9py3.5.yml

[UPDATE] Install tensorbayes

Please note that tensorbayes 0.4.0 is out of date. Please copy a newer version to the env folder (tf1.9py3.5) using tensorbayes.tar

source activate tf1.9py3.5
pip install tensorbayes
tar -xvf tensorbayes.tar
cp -rf /tensorbayes/* /opt/conda/envs/tf1.9py3.5/lib/python3.5/site-packages/tensorbayes/

A.2. Datasets

Please download Office-31 here and unzip extracted features in the datasets folder.

B. Training

We first navigate to model folder, and then run run_lamda.py file as bellow:

cd model

1. A --> W task

python run_lamda.py 1 amazon webcam format csv num_iters 20000 summary_freq 400 learning_rate 0.0001 inorm True batch_size 310 src_class_trade_off 1.0 domain_trade_off 0.1 src_vat_trade_off 0.1 trg_trade_off 0.1 save_grads False cast_data False cnn_size small update_target_loss False m_on_D_trade_off 1.0 m_plus_1_on_D_trade_off 1.0 m_plus_1_on_G_trade_off 1.0 m_on_G_trade_off 0.1 data_path ""

2. A --> D task

python run_lamda.py 1 amazon dslr format csv num_iters 20000 summary_freq 400 learning_rate 0.0001 inorm True batch_size 310 src_class_trade_off 1.0 domain_trade_off 0.1 src_vat_trade_off 1.0 trg_trade_off 0.1 save_grads False cast_data False cnn_size small update_target_loss False m_on_D_trade_off 1.0 m_plus_1_on_D_trade_off 1.0 m_plus_1_on_G_trade_off 1.0 m_on_G_trade_off 0.05 data_path ""

3. D --> W task

python run_lamda.py 1 dslr webcam format csv num_iters 20000 summary_freq 400 learning_rate 0.0001 inorm True batch_size 155 src_class_trade_off 1.0 domain_trade_off 0.1 src_vat_trade_off 0.1 trg_trade_off 0.1 save_grads False cast_data False cnn_size small update_target_loss False m_on_D_trade_off 1.0 m_plus_1_on_D_trade_off 1.0 m_plus_1_on_G_trade_off 1.0 m_on_G_trade_off 0.1 data_path ""

4. W --> D task

python run_lamda.py 1 webcam dslr format csv num_iters 20000 summary_freq 400 learning_rate 0.0001 inorm True batch_size 310 src_class_trade_off 1.0 domain_trade_off 0.1 src_vat_trade_off 0.1 trg_trade_off 0.1 save_grads False cast_data False cnn_size small update_target_loss False m_on_D_trade_off 1.0 m_plus_1_on_D_trade_off 1.0 m_plus_1_on_G_trade_off 1.0 m_on_G_trade_off 0.1 data_path ""

5. D --> A task

python run_lamda.py 1 dslr amazon format csv num_iters 20000  sumary_freq 400 learning_rate 0.0001 inorm True batch_size 155 src_class_trade_off 1.0 domain_trade_off 0.1 src_vat_trade_off 1.0 trg_trade_off 0.1 save_grads False cast_data False cnn_size small update_target_loss False m_on_D_trade_off 1.0 m_plus_1_on_D_trade_off 1.0 m_plus_1_on_G_trade_off 1.0 m_on_G_trade_off 1.0 data_path ""

6. W --> A task

python run_lamda.py 1 webcam amazon format csv num_iters 20000 summary_freq 400 learning_rate 0.0001 inorm True batch_size 310 src_class_trade_off 1.0 domain_trade_off 0.1 src_vat_trade_off 1.0 trg_trade_off 0.1 save_grads False cast_data False cnn_size small update_target_loss False m_on_D_trade_off 1.0 m_plus_1_on_D_trade_off 1.0 m_plus_1_on_G_trade_off 1.0 m_on_G_trade_off 1.0 data_path ""

C. Results

D. Citations

Please cite the paper if LAMDA is helpful for your research:

@InProceedings{pmlr-v139-le21a,
  title = 	 {LAMDA: Label Matching Deep Domain Adaptation},
  author =       {Le, Trung and Nguyen, Tuan and Ho, Nhat and Bui, Hung and Phung, Dinh},
  booktitle = 	 {Proceedings of the 38th International Conference on Machine Learning},
  pages = 	 {6043--6054},
  year = 	 {2021},
  editor = 	 {Meila, Marina and Zhang, Tong},
  volume = 	 {139},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {18--24 Jul},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v139/le21a/le21a.pdf},
  url = 	 {https://proceedings.mlr.press/v139/le21a.html},
  abstract = 	 {Deep domain adaptation (DDA) approaches have recently been shown to perform better than their shallow rivals with better modeling capacity on complex domains (e.g., image, structural data, and sequential data). The underlying idea is to learn domain invariant representations on a latent space that can bridge the gap between source and target domains. Several theoretical studies have established insightful understanding and the benefit of learning domain invariant features; however, they are usually limited to the case where there is no label shift, hence hindering its applicability. In this paper, we propose and study a new challenging setting that allows us to use a Wasserstein distance (WS) to not only quantify the data shift but also to define the label shift directly. We further develop a theory to demonstrate that minimizing the WS of the data shift leads to closing the gap between the source and target data distributions on the latent space (e.g., an intermediate layer of a deep net), while still being able to quantify the label shift with respect to this latent space. Interestingly, our theory can consequently explain certain drawbacks of learning domain invariant features on the latent space. Finally, grounded on the results and guidance of our developed theory, we propose the Label Matching Deep Domain Adaptation (LAMDA) approach that outperforms baselines on real-world datasets for DA problems.}
}

E. References

E.1. Baselines:

[1] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016.

[2] B. Sun and K. Saenko. Deep coral: Correlation alignment for deep domain adaptation. In Gang Hua and Hervé Jéegou, editors, Computer Vision – ECCV 2016 Workshops, pages 443–450, Cham, 2016. Springer International Publishing.

[3] Y. Ganin, E. Ustinova, H. Ajakan, P. Germain, H. Larochelle, F. Laviolette, M. Marchand, and V. Lempitsky. Domain-adversarial training of neural networks. J. Mach. Learn. Res., 17(1):2096–2030, jan 2016.

[4] E. Tzeng, J. Hoffman, K. Saenko, and T. Darrell. Adversarial discriminative domain adaptation. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2962–2971, 2017.

[5] M. Long, Z. Cao, J. Wang, and M. I. Jordan. Conditional adversarial domain adaptation. In Advances in Neural Information Processing Systems 31, pages 1640–1650. Curran Associates, Inc., 2018.

[6] Y. Pan, T. Yao, Y. Li, Y. Wang, C. Ngo, and T. Mei. Transferrable prototypical networks for unsupervised domain adaptation. In CVPR, pages 2234–2242, 2019.

[7] B. B. Damodaran, B. Kellenberger, R. Flamary, D. Tuia, and N. Courty. Deepjdot: Deep joint distribution optimal transport for unsupervised domain adaptation. In Computer Vision - ECCV 2018, pages 467–483. Springer, 2018.

[8] R. Xu, P. Liu, L. Wang, C. Chen, and J. Wang. Reliable weighted optimal transport for unsupervised domain adaptation. In CVPR 2020, June 2020.

E.2. GitHub repositories:

• Some parts of our code (e.g., VAT, evaluation, …) are rewritten with modifications from DIRT-T.