Time series forecasting using massively parallel genetic programming

Abstract

In this paper we propose a massively parallel GP model in hardware as an efficient, flexible and scaleable machine learning system. This fine-grained diffusion architecture consists of a large amount of independent processing nodes that evolve a large number of small, overlapping subpopulations. Every node has an embedded CPU that executes a linear machine code GP representation at a rate of up to 20,000 generations per second. Besides being efficient, implementing the system in VLSI makes it highly portable and makes it possible to target mobile, on-line applications. The SIMD-like architecture also makes the system scalable so that larger problems can be addressed with a system with more processing nodes. Finally, the use of GP representation and VHDL modeling makes the system highly flexible and easy to adapt to different applications. We demonstrate the effectiveness of the system on a time series forecasting application.