Model-SL is an innovative project that implements Split Learning with adaptive encryption and decryption models. These models are parameterized and run locally on user devices, leveraging user history data to train and fine-tune their parameters, enabling personalized and robust encryption-decryption mechanisms. This ensures data security and privacy while enhancing collaborative learning capabilities.
- Localized Encryption and Decryption Models: Each user has locally parameterized encryption and decryption models, ensuring data never leaves their device unencrypted.
- User History-Driven Training: User history data is securely used to train and adapt the encryption and decryption models for personalized performance.
- Split Learning: The neural network is partitioned between the user and server, minimizing the exposure of sensitive data.
- Privacy-Preserving Architecture: The design ensures that raw user data remains secure and never directly shared with the server.
- Modular Design: Future-ready for adding new features and extending encryption mechanisms.
To get started, clone the repository with submodules:
git clone --recurse-submodules https://github.com/AIerLab/model-sl.gitFor deploying in a cloud environment:
bash run_cloud.shFor testing locally on your machine:
bash run_local.sh-
Data Partitioning:
- User data is retained locally, and only encrypted representations are shared with the server.
-
Training Local Models:
- Encryption Model: Trains on user history to generate secure representations of intermediate outputs.
- Decryption Model: Trains in tandem to decode the encrypted data on the server side.
-
Split Learning Integration:
- The encrypted intermediate data is transmitted to the server for further processing, while local models adapt based on user-specific requirements.
-
Continuous Learning:
- User models evolve by leveraging historical data, enhancing encryption robustness over time.
/model-sl/: Main repository containing the split learning implementation./submodules/: Includes experimental modules for encryption and decryption./user-history/: Placeholder for user history data (ensure compliance with privacy standards).run_cloud.sh: Script for cloud execution.run_local.sh: Script for local execution.
- Implementing basic encryption and decryption models.
- Establishing secure communication protocols.
- Testing integration of user-specific data with model parameterization.
- Fine-tuning encryption-decryption models using federated learning techniques.
- Enhancing performance through model optimization and hyperparameter tuning.
- Expanding support for additional use cases and datasets.
As the project is still in its early stages, contributions are welcome!
- Fork the repository.
- Create a feature branch:
git checkout -b feature-name. - Commit your changes:
git commit -m 'Add new feature'. - Push the branch:
git push origin feature-name. - Open a pull request on GitHub.
This project is licensed under the MIT License. See the LICENSE file for more details.
For questions, feature requests, or bug reports, please reach out via GitHub Issues or email us at [email protected].
Stay tuned as we build the future of adaptive, secure split learning!