On Evolutionary Approximation of Sigmoid Function for HW/SW Embedded Systems

Providing machine learningcapabilities on low cost electronic devices is a challenging goal especially inthe context of the Internet of Things paradigm. In order to deliver highperformance machine intelligence on low power devices, suitable hardwareaccelerators have to be introduced. In this paper, we developed a methodenabling to evolve a hardware implementation together with a correspondingsoftware controller for key components of smart embedded systems. The proposed approachis based on a multi-objective design space exploration conducted by means ofextended linear genetic programming. The approach was evaluated in the task ofapproximate sigmoid function design which is an important component of hardwareimplementations of neural networks. During these experiments, we automaticallyre-discovered some approximate sigmoid functions known from the literature. Themethod was implemented as an extension of an existing platform supportingconcurrent evolution of hardware and software of embedded systems.