PaKD Q-Model Compression

Project: Multi-stage Compression of Deep Neural Networks through Pruning, Knowledge Distillation, and Quantization
Lead Professor: Dr. Pooyan Jamshidi
Project lead: Blake Edwards
Contributers: Blake Edwards, Shahriar Iqbal, Stephen Baione

Table of Contents

1. Overview
2. Dependencies
3. Pruning
4. Knowledge Distillation
5. Project Organization
6. Getting Started
7. How to Contribute
8. References

1. Overview

Exploration and analysis of deep neural network knowledge distillation techniques: teacher assisted knowledge distillation, mulitstage knowledge distillation, and early stopping knowledge distillation.

2. Dependencies

Software

Hardware

Minimum memory requirement: 3GB of RAM

3. Pruning

4. Knowledge Distillation

overview of implemented knowledge distillation (KD) methods

5. Project Organization

---

├── LICENSE
├── Makefile           <- Makefile with commands like `make data` or `make train`
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── docs               <- A default Sphinx project; see sphinx-doc.org for details
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
│                         the creator's initials, and a short `-` delimited description, e.g.
│                         `1.0-jqp-initial-data-exploration`.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
│                         generated with `pip freeze > requirements.txt`
│
├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
├── src                <- Source code for use in this project.
│   ├── __init__.py    <- Makes src a Python module
│   │
│   ├── data           <- Scripts to download or generate data
│   │   └── make_dataset.py
│   │
│   ├── features       <- Scripts to turn raw data into features for modeling
│   │   └── build_features.py
│   │
│   ├── models         <- Scripts to train models and then use trained models to make
│   │   │                 predictions
│   │   ├── predict_model.py
│   │   └── train_model.py
│   │
│   └── visualization  <- Scripts to create exploratory and results oriented visualizations
│       └── visualize.py
│
└── tox.ini            <- tox file with settings for running tox; see tox.testrun.org

---

6. Getting Started

7. How to Contribute

8. References

References

1. Asif, U., Tang, J., & Harrer, S. (2019). Ensemble knowledge distillation for learning improved and efficient networks. ArXiv:1909.08097 [Cs]. Retrieved from http://arxiv.org/abs/1909.08097

2. Cheng, Y., Wang, D., Zhou, P., & Zhang, T. (2017). A Survey of Model Compression and Acceleration for Deep Neural Networks. ArXiv, abs/1710.09282.

3. Cole, Casey & Janos, Bethany & Anshari, Dien & Thrasher, James & Strayer, Scott & Valafar, Homayoun. (2016). Recognition of Smoking Gesture Using Smart Watch Technology.

4. Han, S., Mao, H., & Dally, W.J. (2015). Deep Compression: Compressing Deep Neural Network with Pruning, Trained Quantization and Huffman Coding. CoRR, abs/1510.00149.

5. He, Y., Kang, G., Dong, X., Fu, Y., & Yang, Y. (2018). Soft Filter Pruning for Accelerating Deep Convolutional Neural Networks. IJCAI.

6. Hinton, G., Vinyals, O. & Dean, J. (2015). Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531.

7. Iandola, F. N., Han, S., Moskewicz, M. W., Ashraf, K., Dally, W. J. & Keutzer, K. (2017). SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size (cite arxiv:1602.07360Comment: In ICLR Format)

8. Iqbal, M. S., Kotthoff, L., & Jamshidi, P. (2019). Transfer learning for performance modeling of deep neural network systems. ArXiv:1904.02838 [Cs]. Retrieved from http://arxiv.org/abs/1904.02838

9. Lecun, Y., Denker, J. S., Solla, S. A., Howard, R. E., & Jackel, L. D. (1990). Optimal brain damage. In D. Touretzky (Ed.), Advances in Neural Information Processing Systems (NIPS 1989), Denver, CO (Vol. 2). Morgan Kaufmann.

10. Mirzadeh, S., Farajtabar, M., Li, A., & Ghasemzadeh, H. (2019). Improved Knowledge Distillation via Teacher Assistant: Bridging the Gap Between Student and Teacher. ArXiv, abs/1902.03393.

11. Jin, S., Di, S., Liang, X., Tian, J., Tao, D., & Cappello, F. (2019). Deepsz: A novel framework to compress deep neural networks by using error-bounded lossy compression. Proceedings of the 28th International Symposium on High-Performance Parallel and Distributed Computing - HPDC ’19, 159–170. https://doi.org/10.1145/3307681.3326608

12. Singh, P., Verma, V. K., Rai, P., & Namboodiri, V. P. (2019). Play and prune: Adaptive filter pruning for deep model compression. ArXiv:1905.04446 [Cs]. Retrieved from http://arxiv.org/abs/1905.04446

13. Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I.J., & Fergus, R. (2013). Intriguing properties of neural networks. CoRR, abs/1312.6199.

14. You, Z., Yan, K., Ye, J., Ma, M., & Wang, P. (2019). Gate decorator: Global filter pruning method for accelerating deep convolutional neural networks. ArXiv:1909.08174 [Cs, Eess]. Retrieved from http://arxiv.org/abs/1909.08174