Garnett Carl Wilson

Accelerating Genetic Programming through Graphics Processing Units

Graphics Processing Units (GPUs) are in the process of becoming a major source of computational power for numerical applications. Originally designed for application of time-consuming graphics operations, GPUs are stream processors that implement the SIMD paradigm. The true degree of parallelism of GPUs is often hidden from the user, making programming even more flexible and convenient. In this chapter we survey Genetic Programming methods currently ported

Resource Review: Three Open Source Systems for Evolving Programs–Lilgp, ECJ and Grammatical Evolution

We consider three open source codes that have the potential to provide the basis for research into evolutionary program code generation approaches to machine learning. Earlier reviews of information and tools for genetic programming, have considered online resources for ‘‘getting started’’ [1] and a commercial implementation of genetic programming (Discipulus) [2]. The following three systems were judged sufficiently well established for review,

Discovery of email communication networks from the Enron corpus with a genetic algorithm using social network analysis

During the legal investigation of Enron Corporation, the U.S. Federal Regulatory Commission (FERC) made public a substantial data set of the companys internal corporate emails. This work presents a genetic algorithm (GA) approach to social network analysis (SNA) using the Enron corpus. Three SNA metrics, degree, density, and proximity prestige, were applied to the detection of networks with high email activity and presence of important actors with respect to email transactions. Quantitative analysis revealed that density and proximity prestige captured networks of high activity more so than degree. Subsequent qualitative analysis indicated that there were trade-offs in the selection of SNA metrics. Examination of the discovered social networks showed that density and proximity prestige isolated networks involving key actors to a greater extent than degree. In particular, density picked out interesting patterns in terms of email volume, while proximity prestige better isolated key actors at Enron. The roles of the particular actors picked out by the networks as reasons for their prominence are also discussed.

Linear genetic programming GPGPU on Microsoft’s Xbox 360

We describe how to harness the graphics processing abilities of a consumer video game console (Xbox 360) for general programming on graphics processing unit (GPGPU) purposes. In particular, we implement a linear GP (LGP) system to solve classification and regression problems. We conduct inter- and intra-platform benchmarking of the Xbox 360 and PC, using GPU and CPU implementations on both architectures. Platform benchmarking confirms highly integrated CPU and GPU programming flexibility of the Xbox 360, having the potential to alleviate typical GPGPU decisions of allocating particular functionalities to CPU or GPU.

Introducing probabilistic adaptive mapping developmental genetic programming with redundant mappings

Developmental Genetic Programming (DGP) algorithms have explicitly required the search space for a problem to be divided into genotypes and corresponding phenotypes. The two search spaces are often connected with a genotype-phenotype mapping (GPM) intended to model the biological genetic code, where current implementations of this concept involve evolution of the mappings along with evolution of the genotype solutions. This work presents the Probabilistic Adaptive Mapping DGP (PAM DGP), a new developmental implementation that involves research contributions in the areas of GPMs and coevolution. The algorithm component of PAM DGP is demonstrated to overcome coevolutionary performance problems that are identified and empirically benchmarked against the latest competing algorithm that adapts similar GPMs. An adaptive redundant mapping encoding is then incorporated into PAM DGP for further performance enhancement. PAM DGP with two mapping types are compared to the competing Adaptive Mapping algorithm and Traditional GP in two medical classification domains, where PAM DGP with redundant encodings is found to provide superior fitness performance over the other algorithms through it’s ability to explicitly decrease the size of the function set during evolution.

Deployment of CPU and GPU-based genetic programming on heterogeneous devices

A widely available and economic means of increasing the computing power applied to a problem is to use modern graphics processing units (GPUs) for parallel processing. We present a new, optimized general methodology for deploying genetic programming (GP) to the PC, Xbox 360 video game console, and Zune portable media device. This work describes, for the first time, the implementation considerations necessary to maximize available CPU and GPU (where available) usage on the three separate hardware platforms. We demonstrate the first instance of GP using portable digital media device hardware. The work also presents, for the first time, an Xbox 360 implementation that uses the GPU for fitness evaluation. Implementations on each platform are also benchmarked on the basis of execution time for an established GP regression benchmark.

A comparison of Cartesian genetic programming and linear genetic programming

Two prominent genetic programming approaches are the graph-based Cartesian Genetic Programming (CGP) and Linear Genetic Programming (LGP). Recently, a formal algorithm for constructing a directed acyclic graph (DAG) from a classical LGP instruction sequence has been established. Given graph-based LGP and traditional CGP, this paper investigates the similarities and differences between the two implementations, and establishes that the significant difference between them is each algorithms means of restricting interconnectivity of nodes. The work then goes on to compare the performance of two representations each (with varied connectivity) of LGP and CGP to a directed cyclic graph (DCG) GP with no connectivity restrictions on a medical classification and regression benchmark.

Exploring geo-temporal differences using GTdiff

Many data sets exist that contain both geospatial and temporal elements, in addition to the core data that requires analysis. Within such data sets, it can be difficult to determine how the data have changed over spatial and temporal ranges. In this design study we present a system for dynamically exploring geo-temporal changes in the data. GTdiff provides a visual approach to representing differences in the data within user-defined spatial and temporal limits, illustrating when and where increases and/or decreases have occurred. The system makes extensive use of spatial and temporal filtering and binning, geo-visualization, colour encoding, and multiple coordinated views. It is highly interactive, supporting knowledge discovery through exploration and analysis of the data. A case study is presented illustrating the benefits of using GTdiff to analyze the changes in the catch data of the cod fisheries off the coast of Newfoundland, Canada from 1948 to 2006.

Prediction of Interday Stock Prices Using Developmental and Linear Genetic Programming

A developmental co-evolutionary genetic programming approach (PAM DGP) is compared to a standard linear genetic programming (LGP) implementation for trading of stocks across market sectors. Both implementations were found to be impressively robust to market fluctuations while reacting efficiently to opportunities for profit, where PAM DGP proved slightly more reactive to market changes than LGP. PAM DGP outperformed, or was competitive with, LGP for all stocks tested. Both implementations had very impressive accuracy in choosing both profitable buy trades and sells that prevented losses, where this occurred in the context of moderately active trading for all stocks. The algorithms also appropriately maintained maximal investment in order to profit from sustained market upswings.

Interday foreign exchange trading using linear genetic programming

Foreign exchange (forex) market trading using evolutionary algorithms is an active and controversial area of research. We investigate the use of a linear genetic programming (LGP) system for automated forex trading of four major currency pairs. Fitness functions with varying degrees of conservatism through the incorporation of maximum drawdown are considered. The use of the fitness types in the LGP system for different currency value trends are examined in terms of performance over time, underlying trading strategies, and overall profitability. An analysis of trade profitability shows that the LGP system is very accurate at both buying to achieve profit and selling to prevent loss, with moderate levels of trading activity.