"Team to Present Work Around Spiking Neural Networks at DATE Conference"
A research team from the College of Engineering at the University of Missouri led by assistant professor Khaza Anuarul Hoque will present two papers at the Design Automation and Test in Europe Conference and Exhibition (DATE). Both publications advance research into Spiking Neural Networks (SNNs), a type of Machine Learning (ML) that identifies objects using binary code, which is the "0" and "1" system employed in digital computation. Hoque, director of the Dependable Cyber-Physical Systems (DCPS) Laboratory, noted that ML hardware is highly power-hungry. SNNs are brain-inspired and energy-efficient compared to other deep neural networks, which is one of the reasons for their growing popularity. Ph.D. student Ayesha Siddique will present her paper on enhancing the reliability of SNNs used in hardware. Syed Tihaam Ahmad, a Ph.D. student, will present his paper on protecting these systems from cyber threats. Although SNNs are already energy-efficient, approximation approaches cut energy requirements even more at the expense of a little accuracy. However, doing so raises the vulnerability of a system to malicious activity. Ahmad proposes a design methodology to produce a more secure version of SNNs that involves two innovative defense techniques: precision scaling and approximate quantization-aware filtering. According to Ahmad, both methods significantly improve the robustness of approximate SNNs. This article continues to discuss the research on improving the reliability and security of SNNs.