Ransomware Detection Using Limited Precision Deep Learning Structure in FPGA

Abstract

We propose a fast ransomware detection method using Memory-Assisted-Stochastic-Dynamic-Fixed-Point arithmetic using a four-layer Deep Belief Network (DBN) structure. The method stores random bit-streams in memory to produce efficient cross-correlation for the stochastic computation in FPGA. The memory technique for stochastic computation with dynamic fixed-point arithmetic trains the Deep Belief Network (DBN) to detect ransomwares with 91% precision rate and detection speed of.006ms. The method represents a promising step toward improving ransomware detection in devices with limited power and memory resources such as the Internet of Things (IoTs).