STAM Center Researchers Present Privacy-Preserving SNNs Framework at ACNS

The 24th International Conference on Applied Cryptography and Network Security (ACNS 2026) was held between the 22nd and 25th of June 2026, at Stony Brook University in New York. As one of the premier international conferences in applied cryptography, cybersecurity, and privacy, ACNS brings together researchers from around the world to present advances in secure computing, applied cryptography, cybersecurity, and privacy. This year, we are happy to share that researchers from the center showcased their latest work in privacy-preserving artificial intelligence at this conference

Ph.D. Candidate Nges Brian Njungle presented the paper, “PrivSpike: A Privacy-Preserving Framework for Deep Spiking Neural Networks Using Homomorphic Encryption,” introducing new techniques for enabling secure inference in deep Spiking Neural Networks (SNNs).

Deep learning has transformed modern artificial intelligence, driving applications in healthcare, finance, autonomous systems, and cybersecurity. However, these advances often come at the cost of requiring access to large volumes of sensitive data, making privacy a critical concern. Spiking Neural Networks have emerged as an energy-efficient alternative to conventional deep neural networks by more closely modeling the behavior of biological neurons. They are particularly well suited for processing neuromorphic, spatiotemporal, and event-driven data, where they achieve superior efficiency and performance over existing machine learning methods. Like traditional deep learning models, however, SNNs also rely on sensitive data and therefore face many of the same privacy challenges.

To address these challenges, researchers from the AITS Lab and the ASCS Lab developed PrivSpike, a framework that combines homomorphic encryption with deep Spiking Neural Networks, enabling inference to be performed directly on encrypted data without revealing the underlying information.

PrivSpike introduces two novel algorithms for efficiently evaluating SNNs in the encrypted domain, significantly improving the practicality of encrypted SNN inference, by delivering performance of up to 50× improvement over previously reported encrypted approaches. read paper Here

PrivSpike advances the development of secure AI systems that protect sensitive information without sacrificing computational capability. The framework represents an important step toward deploying privacy-preserving AI in applications where both security and performance are essential.

The research was primarily lead by Nges Brian Njungle, Eric Jahns, Milan Stojkov, and Prof. Michel A. Kinsy as part of the STAM Center’s ongoing efforts to develop secure, trustworthy, and privacy-preserving computing systems. The work reflects the center’s mission to advance technologies that strengthen the security, resilience, and trustworthiness of next-generation intelligent systems.