PETGNNs

Inspiration In today’s digital age, privacy concerns are more prevalent than ever. The idea behind PPFGNN came from the need to develop advanced machine learning models that not only deliver high accuracy but also ensure the privacy of sensitive data. We were inspired by the growing demand for secure and reliable AI solutions, particularly in areas like fake news detection and fraud prevention.

What it does PPFGNN is a Graph Neural Network (GNN) project designed to tackle fake news and fraud detection. It employs privacy-preserving techniques such as differential privacy, secure multi-party computation (SMPC), and homomorphic encryption. The system allows multiple collaborators to upload their datasets securely, train models collectively, and perform incremental and decremental learning while ensuring data privacy and security throughout the process.

How we built it We built PPFGNN using a combination of cutting-edge technologies. The frontend was developed using React Vite, while the backend is powered by Express.js. For the GNN and cryptographic operations, we utilized TensorFlow Federated and TikTok's PETACE framework. The project architecture allows for both client-side data validation and server-side advanced processing. We also integrated continuous validation and testing to maintain model quality.

Challenges we ran into One of the major challenges was ensuring seamless integration of various privacy-preserving techniques without compromising on performance. We also faced difficulties with CUDA compatibility issues, which required us to explore alternative methods for speeding up computations. Coordinating multiple collaborators and managing the incremental and decremental training strategies were also complex tasks that required meticulous planning and execution.

Accomplishments that we're proud of We’re proud of successfully developing a system that balances high-performance machine learning with robust privacy protections. Implementing a user-friendly UI that allows for easy model training, visualization, and management was a significant achievement. Additionally, overcoming technical challenges and integrating multiple advanced privacy-preserving methods into a cohesive system was a milestone we’re particularly proud of.

What we learned Throughout this project, we learned a great deal about the practical implementation of privacy-preserving technologies in machine learning. We gained insights into the intricacies of GNNs and how to efficiently handle large, collaborative datasets. This experience also enhanced our understanding of the importance of continuous validation and testing in maintaining model integrity.

What's next for PPFGNN Moving forward, we plan to refine and expand PPFGNN’s capabilities. We aim to enhance the user interface further, making it even more intuitive and accessible. We’re also looking into integrating additional privacy-preserving techniques and improving the scalability of the system. Our goal is to deploy PPFGNN in real-world applications, providing organizations with a powerful tool for secure and effective data analysis.

Built With

• express.js
• flask
• kubernetes
• react-vite
• tensorflow-federated