Abstract. A Single Instruction Multiple Thread CUDA interpreter provides SIMD like parallel evaluation of the whole GP population of 1 4 million reverse polish notation (RPN) expressions on graphics cards and nVidia Tesla. Using sub-machine code tree GP a sustain peak performance of 665 billion GP operations per second (10,000 speed up) and an average of 22 peta GP ops per day is reported for a single GPU card on a Boolean induction benchmark never attempted before, let alone solved. 1

Spectra-Based Fault Localisation (SBFL) aims to assist de- bugging by applying risk evaluation formulæ (sometimes called suspiciousness metrics) to program spectra and ranking statements according to the predicted risk. Designing a risk evaluation formula is often an intuitive process done by human software engineer. This paper presents a Genetic Programming approach for evolving risk assessment formulæ. The empirical evaluation The Emerald supercomputer contains more than a thousand CPU cores and several hundred nVidia Tesla. A genetic programming GeneChip datamining application which searches for non-linear gene based prediction of long term survival following breast cancer surgery was transferred without change and run on part of the Emerald cluster. At 7 giga. GP-evolved GPopS equations can consistently outperform many of the human-designed formulæ, such as Tarantula, Ochiai, Jaccard, Ample, and Wong1/2, up to 5.9 times. More importantly, they can perform equally as well as Op2, which was recently proved to be optimal against If-Then-Else-2 (ITE2) structure, or even outperform it against other program structures. −1 it is the fastest ever genetic programming application of this type.

Abstract — This work examines a novel method that provides a parallel search of a very large network space consisting of fisheries management data. The parallel search solution is capable of determining global maxima of the search space using brute force search, compared to local optima located by machine learning solutions such as evolutionary computation. The actual solutions from the best machine learning technique, called Probabilistic Adaptive Mapping Developmental Genetic Algorithm, are compared by a fisheries expert to the global maxima solutions returned by parallel search. In addition, the time required for parallel search, for both CPU and GPU-optimized solutions, are compared to those required for machine learning solutions. The GPU parallel computing solution was found to have a speedup of over 10,000x, in excess of most similar performance comparison studies in the literature. An expert found that overall the machine learning solutions pro-duced more interesting results by locating local optima than global optima determined by parallel processing. I.

Spectra-Based Fault Localisation (SBFL) aims to assist de- bugging by applying risk evaluation formulæ (sometimes called suspiciousness metrics) to program spectra and ranking statements according to the predicted risk. Designing a risk evaluation formula is often an intuitive process done by human software engineer. This paper presents a Genetic Programming approach for evolving risk assessment formulæ. The empirical evaluation There has been much recent interest in genetic improvement of programs. However, genetic improvement has yet to scale beyond toy laboratory programs. We seek to overcome this scalability barrier. We evolved a widely-used and highly complex 50 000 line GP-evolved system, seeking equations can consistently improved versions outperform that are many faster of than the the human-designed original, yet at least formulæ, as good semantically. such as Tarantula,

Spectra-Based Fault Localisation (SBFL) aims to assist de- bugging by applying risk evaluation formulæ (sometimes called suspiciousness metrics) to program spectra and ranking statements according to the predicted risk. Designing a risk evaluation formula is often an intuitive process done by human software engineer. This paper presents a Genetic Programming approach for evolving risk assessment formulæ. The empirical evaluation using 92 faults from four Unix utilities produces promising results1. GP-evolved equations can consistently outperform many of the human-designed formulæ, such as Tarantula, Ochiai, Jaccard, Ample, and Wong1/2, up to 5.9 times. More importantly, they can perform equally as well as Op2, which was recently proved to be optimal against If-Then-Else-2 (ITE2) structure, or even outperform it against other program structures. 1 The program spectra data used in the paper, as well as the complete empirical results, are available from: