Federated Learning Mammogram Simulation
Budget: ₹12,500 – ₹37,500 INR
I need a straightforward, well-commented Python prototype that shows how federated learning can be applied to healthcare data processing—more precisely, to medical image analysis on mammograms.
Scope
• Build a minimal yet functional simulation where several virtual clients train a shared convolutional model on local mammogram data while keeping the raw images private.
• Use an open-source framework that already supports federated workflows—TensorFlow Federated, Flower, or PySyft—whichever lets you move fastest.
• Include synthetic or public mammogram samples so the code runs out-of-the-box (no licensed datasets required).
Core Requirements
1. Clear project structure with a server-client loop, model aggregation, and at least three simulated clients.
2. A small CNN suitable for image classification; accuracy reporting after each aggregation round.
3. Comments explaining the key privacy-preserving steps and any assumptions.
4. A short README with setup steps, how to launch the simulation, and where to plug in real data later.
Deliverables
• All source code and requirements.txt/conda-environment.yml.
• The README and a brief note on possible next steps (e.g., adding differential privacy or secure aggregation).
I’m aiming for a concise, functional demo rather than a full production system, so focus on getting the essential pieces working cleanly and quickly.
Scope
• Build a minimal yet functional simulation where several virtual clients train a shared convolutional model on local mammogram data while keeping the raw images private.
• Use an open-source framework that already supports federated workflows—TensorFlow Federated, Flower, or PySyft—whichever lets you move fastest.
• Include synthetic or public mammogram samples so the code runs out-of-the-box (no licensed datasets required).
Core Requirements
1. Clear project structure with a server-client loop, model aggregation, and at least three simulated clients.
2. A small CNN suitable for image classification; accuracy reporting after each aggregation round.
3. Comments explaining the key privacy-preserving steps and any assumptions.
4. A short README with setup steps, how to launch the simulation, and where to plug in real data later.
Deliverables
• All source code and requirements.txt/conda-environment.yml.
• The README and a brief note on possible next steps (e.g., adding differential privacy or secure aggregation).
I’m aiming for a concise, functional demo rather than a full production system, so focus on getting the essential pieces working cleanly and quickly.