Efficient ensemble machine learning implementation on FPGA using partial reconfiguration

Abstract

Ensemble Machine Learning (EML) consists of the combination of multiple Artificial Intelligence algorithms. This paper presents an efficient FPGA implementation of an Ensemble based on Long Short-Term Memory Networks (LSTM). For an efficient implementation, the proposed design uses the Partial Reconfiguration function available for FPGAs. Results are presented in terms of resources utilization, reconfiguration speed, power consumption and maximum clock frequency.