Python SNN with Homomorphic Encryption
Budget: ₹12,500 – ₹37,500 INR
I’m running an R-and-D exploration into privacy-preserving spiking neural networks and need a compact proof-of-concept written in Python. The core idea is to take a small SNN model, encrypt its inputs (and ideally weights or activations) with a well-known fully homomorphic encryption (FHE) scheme, run inference while data remains encrypted, then decrypt the result for evaluation.
You’re free to choose the specific Python stack, but I expect something mainstream and well-documented—e.g., Brian2, BindsNET, Lava, or a minimal NumPy/PyTorch implementation on the neural side, paired with Microsoft SEAL, Pyfhel, or a similar library for FHE. Keep external dependencies to a minimum unless a package is essential to demonstrating the concept.
Deliverables
• Clean, runnable Python code (notebooks or scripts) that:
– Builds or loads a simple spiking neural network,
– Encrypts sample inputs, performs encrypted inference, and decrypts outputs.
• A short README describing setup, command-line usage, and library versions.
• Brief technical note (PDF or Markdown) explaining design choices, limitations, and next-step research opportunities.
Acceptance criteria
• Decrypted outputs match plaintext inference within an acceptable error tolerance you specify up front.
• All code executes end-to-end on a standard CPU environment (no GPUs required).
• No hard-coded paths or proprietary data.
This is experimental work, so clarity and reproducibility matter more than raw speed. If you enjoy pushing the edge of neural computation and cryptography, I’d love to see what you can build.
You’re free to choose the specific Python stack, but I expect something mainstream and well-documented—e.g., Brian2, BindsNET, Lava, or a minimal NumPy/PyTorch implementation on the neural side, paired with Microsoft SEAL, Pyfhel, or a similar library for FHE. Keep external dependencies to a minimum unless a package is essential to demonstrating the concept.
Deliverables
• Clean, runnable Python code (notebooks or scripts) that:
– Builds or loads a simple spiking neural network,
– Encrypts sample inputs, performs encrypted inference, and decrypts outputs.
• A short README describing setup, command-line usage, and library versions.
• Brief technical note (PDF or Markdown) explaining design choices, limitations, and next-step research opportunities.
Acceptance criteria
• Decrypted outputs match plaintext inference within an acceptable error tolerance you specify up front.
• All code executes end-to-end on a standard CPU environment (no GPUs required).
• No hard-coded paths or proprietary data.
This is experimental work, so clarity and reproducibility matter more than raw speed. If you enjoy pushing the edge of neural computation and cryptography, I’d love to see what you can build.