Abstract:With the rapid development of AI technology, we have witnessed numerous innovations and conveniences. However, along with these advancements come privacy threats and risks. Fully Homomorphic Encryption (FHE) emerges as a key technology for privacy-preserving computation, enabling computations while maintaining data privacy. Nevertheless, FHE has limitations in processing continuous non-polynomial functions as it is restricted to discrete integers and supports only addition and multiplication. Spiking Neural Networks (SNNs) operate on discrete spike signals, naturally aligning with the properties of FHE. In this paper, we present a framework called FHE-DiCSNN. This framework is based on the efficient TFHE scheme and leverages the discrete properties of SNNs to achieve high prediction performance on ciphertexts. Firstly, by employing bootstrapping techniques, we successfully implement computations of the Leaky Integrate-and-Fire neuron model on ciphertexts. Through bootstrapping, we can facilitate computations for SNNs of arbitrary depth. This framework can be extended to other spiking neuron models, providing a novel framework for the homomorphic evaluation of SNNs. Secondly, inspired by CNNs, we adopt convolutional methods to replace Poisson encoding. This not only enhances accuracy but also mitigates the issue of prolonged simulation time caused by random encoding. Furthermore, we employ engineering techniques to parallelize the computation of bootstrapping, resulting in a significant improvement in computational efficiency. Finally, we evaluate our model on the MNIST dataset. Experimental results demonstrate that, with the optimal parameter configuration, FHE-DiCSNN achieves an accuracy of 97.94% on ciphertexts, with a loss of only 0.53% compared to the original network's accuracy of 98.47%. Moreover, each prediction requires only 0.75 seconds of computation time

Privacy-preserving Linear Computations in Spiking Neural P Systems

Spiking neural P systems: main ideas and results

Spiking neural P systems with neuron permeability

A Federated Learning Protocol for Spiking Neural Membrane Systems

Spiking Neural P Systems with Delay on Synapses

Parallel Binary Image Cryptosystem Via Spiking Neural Networks Variants

Efficient Generalized Selective Private Function Evaluation With Applications In Biometric Authentication

New Compact Finite-Field Arithmetic Circuits Over GF(p) Based On Spiking Neural P Systems With Communication On Request Implemented in a Low Cost FPGA

Spiking neural P systems with autapses

Privacy-Preserving Discretized Spiking Neural Networks

Efficient Privacy-Preserving Convolutional Spiking Neural Networks with FHE

SecureML: A System for Scalable Privacy-Preserving Machine Learning

BrainLeaks: On the Privacy-Preserving Properties of Neuromorphic Architectures against Model Inversion Attacks

Toward Scalable and Privacy-preserving Deep Neural Network Via Algorithmic-Cryptographic Co-design

Towards Scalable and Privacy-Preserving Deep Neural Network via Algorithmic-Cryptographic Co-design

Asynchronous Numerical Spiking Neural Membrane Systems with Local Synchronization

Robust Stable Spiking Neural Networks

Privacy-Preserving Deep Learning with SPDZ

OPAF: Optimized Secure Two-Party Computation Protocols for Nonlinear Activation Functions in Recurrent Neural Network

Spiking Neural P Systems With Microglia

Enhancing the Robustness of Spiking Neural Networks with Stochastic Gating Mechanisms