A Wisdom of Crowds approach to Supervised Learning Ensembles using FPGAs

In his 2004 book The Wisdom of Crowds, author James Surowiecki details the successes and failures of collective wisdom in Business, Economics, Societies and Nations. The implications of the intelligence of a collective decision of independent judgements are far reaching and include such important things as the success of democracy. Furthermore, the Wisdom of Crowds also finds application in Machine Learning where the collective performance of a group of independent weak learners can outperform individual expert learners. In this paper, we consider a quantitative look at a Wisdom of Crowds approach to create successful supervised learning ensembles. Specifically, this paper considers four (4) characteristics of learners that create a wise ensemble such as diversity of learner, learner independence, decentralization and aggregation of learner results. Design data and equations are provided to scale up a wise ensemble. Equally important, this systems level look at the ensemble provides clarity with respect to when ensemble performance fails. This paper also considers the challenging task of improving the performance of an ensemble of independent learners. This paper also considers implementation of such a supervised learning ensemble using Field Programmable Gate Arrays (FPGAs). The approach described in this paper applies to high performance, high throughput and scalable implementations. Design data is provided for FPGA implementation of a wise ensemble. Finally, an example is provided based on the Xilinx Adaptive Compute Acceleration Platform (ACAP) using the Versal Premium Series device to illustrate the concepts in this paper.