A Homomorphic Encryption Framework for Privacy-Preserving Spiking Neural Networks

Round 1

Reviewer 1 Report

This paper discusses the optimization and security challenges in machine learning and deep neural networks. It explores the use of homomorphic encryption and spiking neural networks. The authors provide an overview of deep learning in neural networks and the application of convolutional neural networks. They also introduce the concept of artificial spiking neural networks and compare different neuromorphic solutions.

However, the paper is not clear in some aspects, and the authors should explain clearly the following issues:

1. What are the optimization and security challenges in machine learning and deep neural networks?

2. What is the difference between artificial spiking neural networks and other neuromorphic solutions?

3. Can homomorphic encryption and spiking neural networks applied to other areas? What are the possible limitations?

n

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper is basically a combination of SNN (using Morse library) and Homomorphic  Encryption (using Pyfhel library). I think the uniqueness of this work is the combination. Other than that, it isn't easy to find the novelty. Following are my comments:

1. It would be better that the authors explain more about HE. How it works and why it maintains the ability to calculate the data. Section 2.3 just briefly introduces HE and I think it is difficult to get the idea.

2. Based on the results, HE introduces some losses in terms of accuracy. Because of that, it is hard to justify the work. Will it be better if we encrypt the data using common techniques like AES and then decode it on the server side? The accuracy will be maintained and potentially some overhead in terms of accuracy.

3. What is the overhead in terms of execution time compared to the non-protected method?

4. I don't think LeNet 5 is considered DNN. What is the result when we apply this framework at deeper networks like VGG or ResNet?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have answered my questions/concerns.