Result: High Accuracy Training of Stochastic Computing-Based Host Intrusion Detection System for Smart Resource-Constrained-IoT.

In this age of the Internet of Things, technology in every sphere of life is getting smarter by the day. However, many resource-constrained IoT devices are thwarting the way. A new hardware accelerator, the Stochastic Computing (SC)-based hardware design, has come to the rescue. The SC-based design has been proven to be capable of accommodating complex processes with little silicon footprint. In this connection, many SC-based deep neural network applications have been presented in recent literature. However, the mismatch between Standard Binary Radix Computing training and SC-based feedforward networks limits the full benefit of SC-based design. To overcome this dilemma, we introduce a novel SC-based deep neural network where the equations of conventional training are derived from the SC hardware module. In this context, we test our network for a Host-based Intrusion Detection System specifically designed for resource-constrained IoT. Both simulation and real network experiments exhibit a high detection rate, virtually the same performance as the standard HIDS designed in a Python environment. [ABSTRACT FROM AUTHOR]