Abstract
Feature extraction and non-linear prediction are standard steps in a time series prediction system. In this talk, some of our recent results in these areas will be described.
The success of machine learning (ML) methods, among other things, depends on a suitable choice of input features which are usually selected by domain-experts. We propose a systematic way for generating suitable features using a context-free grammar. Proven feature selection techniques are used to develop better features in the expanded feature space. Feature generation and selection, based on wavelets and technical indicators, are combined using grammatical evolution where an initial population of well performing individuals are evolved. The proposed method is empirically demonstrated by predicting financial and electricity load time-series using kernel methods.
The kernel recursive least squares (KRLS) algorithm performs non-linear regression in an online manner, with similar computational requirements to linear techniques. An implementation of the KRLS algorithm utilising pipelining and vectorisation for performance; and microcoding for reusability is described. The design can be scaled to allow tradeoffs between capacity, performance and area. Compared with a central processing unit (CPU) and digital signal processor (DSP), the processor improves on execution time, latency and energy consumption by factors of 5, 5 and 12 respectively.
Biography
Philip Leong received the B.Sc., B.E. and Ph.D. degrees from the University of Sydney. He is currently an Associate Professor in the School of Electrical and Information Engineering at the University of Sydney, a Visiting Professor at Imperial College, London and the Chief Technology Advisor to Cluster Technology.
He was the co-founder and program co-chair of the International Conference on Field Programmable Technology (FPT); program co-chair of the International Conference on Field Programmable Logic and Applications (FPL) and is the Associate Editor-in-Chief for the ACM Transactions on Reconfigurable Technology and Systems. The author of more than 100 technical papers and 4 patents, Dr. Leong was the recipient of the 2005 FPT conference Best Paper as well as the 2007 and 2008 FPL conference Stamatis Vassiliadis Outstanding Paper awards.