🔍 American University of Sharjah Machine Learning 504 Generative AI – Group Project: Code Demonstration.
📚 Course: Machine Learning 504
👨‍💻 Project Title: privacy preservation of images using Diffusion model by MachineNotLearning
🎓 Team Members: Affan Murtaza B00104949; Kareem Abed B00106985; Alizar Farhan B00106512

In this video, we demonstrate the full code execution for our group project submission in Machine Learning 504. The project applies core ML concepts such as DDPM using Diffusion models for preservation of Images to solve This work investigates incorporating differential privacy in generative models as a framework for generating useful synthetic data with confidential information protection. We put emphasis on state-of-the-art generators, in particular Generative Adversarial Networks (GANs) and recent diffusion models, as we study the effect of privacy-compliant training procedures on their successful operation. Specifically, we employ Differentially Private Stochastic Gradient Descent (DP-SGD) to train these models and introduce a novel approach named DP-DistillMD that integrates knowledge distillation and differential privacy for improving the training of diffusion-based generative models. This paper investigates the privacy-utility trade-off of these methods: while straightforward DP-SGD leads to a degradation in image quality and model performance, our suggested DP-DistillMD method recovers generation quality drastically under the same privacy constraints. Membership inference attacks conducted on the trained models serve as a way to quantify privacy leakage. The findings indicate that models trained using our privacy mechanism approaches, namely DP-DistillMD, have a high capacity to withstand such attacks and thus reveal better privacy protection. Overall, the findings illustrate that it is feasible to incorporate rigorous privacy protocols into state-of-the-art generative models to achieve synthetic data generation that maintains high utility akin to actual data while satisfying rigorous privacy guarantees.

🚀 What’s Covered in the Video:

Environment setup
Dataset overview and preprocessing
Model training and evaluation
Results and accuracy metrics
Final output and analysis
This demonstration reflects our collaborative effort and understanding of real-world machine learning workflows.

🔗 GitHub Repository :

📌 Note: This video is part of an academic submission and intended for educational purposes.

👍 Don’t forget to like, comment, and subscribe if you found this useful!

#MachineLearning #ML504 #StudentProject #CodeRun #PythonML #DataScience #MachineLearningProject #UniversityLife