Private Inference on Layered Spiking Neural P Systems

Layered Spiking Neural P systems (LSN P systems) are a special class of Spiking Neural Networks (SNNs) used to solve classification problems. These types of networks are inspired directly by the biological brains, and their main advantage is low energy consumption when run on neuromorphic hardware. Since this special type of hardware is not yet available on conventional computers, the most convenient method for using an LSN P system to perform data classification is through a cloud platform. This raises privacy concerns for the users since they expose their data to the cloud provider. This paper presents a new privacy-preserving inference protocol for LSN P systems. The protocol allows one party, called the client, to use a pre-trained LSN P system hosted by another party, called the server, without compromising the privacy of the input or the result. The paper also discusses two brute-force attacks on the protocol and shows that the probability that the server compromises the client’s confidentiality is negligible.