DMCA
Fast genetic programming on GPUs (2007)
Venue: | Proceedings of the 10th European Conference on Genetic Programming, volume 4445 of LNCS |
Citations: | 34 - 8 self |
Citations
553 | A survey of general–purpose computation on graphics hardware
- Owens, Luebke, et al.
(Show Context)
Citation Context ...ematical operations [6]. However, until recently it was cumbersome to use this resource for general purpose computing. For a general survey on algorithms implemented on GPUs the reader is referred to =-=[7]-=-. For example, discrete wavelet transformations [8], the solution of dense linear systems [9], physics simulations for games, fluid simulators [10], etc., have been shown to be executed faster on GPUs... |
331 |
Genetic Programming : An Introduction.
- Banzhaf, Nordin, et al.
- 1998
(Show Context)
Citation Context ...lgorithm, and does not guarantee an increase in performance under all circumstances. In other applications, one could select the number of fitness cases, e.g.. by stochastic sampling or other methods =-=[2]-=-. Should the system need to be tested against a complete input set, however, this approach would not be suitable. Another method involves compiling the evolved expression to executable code or even us... |
269 |
Genetic Programming: On the Programming of Computers by Natural Selection
- Koza
- 1992
(Show Context)
Citation Context ... throughout the run. As the GPU sometimes results in slower performance, we need to verify that on average, there is an advantage. Regression We evolved functions that regressed over x 6 − 2x 4 + x 2 =-=[22]-=-. We tested the evaluation difference using a number of test cases. In each instance, the test cases were uniformly distributed between -1 to +1. We also changed the maximum length of the CGP graph. H... |
230 | Cartesian genetic programming. In:
- Miller, Thomson
- 2000
(Show Context)
Citation Context ...l world problems, rather than on arbitrary expressions. The GP representation we chose to use here is CGP, but similar results should be obtained from other representations. CGP is fully described in =-=[21]-=-. In the benchmark experiments, the expression lengths were uniform throughout the tests. However, in real GP the length of the expressions vary throughout the run. As the GPU sometimes results in slo... |
116 | Accelerator: Using data parallelism to program GPUs for general-purpose uses
- Tarditi, Puri, et al.
- 2006
(Show Context)
Citation Context ...ormance issues compared to C++ and JIT languages such as Java or C♯ 1 . Accelerator: Recently a .Net assembly called Accelerator was released that provides access to the GPU via the DirectX interface =-=[20]-=-. The system is completely abstracted from the GPU, and presents the end user with only arrays that can be operated on in parallel. Unfortunately, the system is only available for the Windows platform... |
81 | LUGPU: Efficient algorithms for solving dense linear systems on graphics hardware
- Galoppo, Govindaraju, et al.
(Show Context)
Citation Context ...eneral purpose computing. For a general survey on algorithms implemented on GPUs the reader is referred to [7]. For example, discrete wavelet transformations [8], the solution of dense linear systems =-=[9]-=-, physics simulations for games, fluid simulators [10], etc., have been shown to be executed faster on GPUs. In this paper we demonstrate a method for using the GPU as an evaluator for genetic program... |
76 | Using modern graphics architectures for general-purpose computing: a framework and analysis. In:
- Thompson, Hahn, et al.
- 2002
(Show Context)
Citation Context ...cifically developed. Recently it has become possible to access the processing power of the graphic processing unit (GPU). Modern GPUs are extremely good at performing parallel mathematical operations =-=[6]-=-. However, until recently it was cumbersome to use this resource for general purpose computing. For a general survey on algorithms implemented on GPUs the reader is referred to [7]. For example, discr... |
63 | On linear genetic programming.
- Brameier
- 2004
(Show Context)
Citation Context ...itable. Another method involves compiling the evolved expression to executable code or even using binary code directly [3]. Writing expressions as native code or in a similar vain has many advantages =-=[4]-=-. The compiler or a hand-written algorithm can perform optimisations, e.g. by removing redundant code, which in addition to directly running the expression gives a significant increase in performance.... |
42 | Efficient Evolution of Machine Code for {CISC} Architectures using Blocks and Homologous Crossover.
- Nordin, Banzhaf, et al.
- 1999
(Show Context)
Citation Context ...to be tested against a complete input set, however, this approach would not be suitable. Another method involves compiling the evolved expression to executable code or even using binary code directly =-=[3]-=-. Writing expressions as native code or in a similar vain has many advantages [4]. The compiler or a hand-written algorithm can perform optimisations, e.g. by removing redundant code, which in additio... |
23 |
Evolving multiplier circuits by training set and training vector partitioning
- Torresen
- 2003
(Show Context)
Citation Context ...onable time we have to also find innovative approaches that let us solve these problems. Traditional GP has difficulty with scaling. For example, the largest evolved multiplier has 1024 fitness cases =-=[24]-=-. In the same time it would take a CPU implementation to evaluate an individual with that many fitness cases, we could test more than 65536 fitness cases on a GPU. This leads to a gap between what we ... |
20 | Parallel Genetic Algorithms on Programmable Graphics Hardware
- Yu, Chen, et al.
- 2005
(Show Context)
Citation Context ...n this, and evaluates expressions on the GPU. There all the operations are treated as graphics operations, which makes implementation difficult and limits the flexibility of the evaluations. Yu et al =-=[13]-=-, on the other hand, implement a Genetic Algorithm on GPUs. Depending on population size, they find a speed up factor of up to 20. Here both the genetic operators and fitness evaluation are performed ... |
20 | Repeated Sequences in Linear Genetic Programming Genomes
- Langdon, Banzhaf
- 2005
(Show Context)
Citation Context ...cases, with larger expression sizes. Classification in Bioinformatics In this experiment we investigate the behaviour on another classification problem, this time a protein classifier as described in =-=[23]-=-. Here the task is to predict the location of a protein in a cell, from the amino acids in the particular protein. We used the entire dataset as the training set. The set consisted of 2427 entries, wi... |
19 | Dynamic subset selection based on a fitness case topology
- Lasarczyk, Dittrich, et al.
- 2004
(Show Context)
Citation Context ...programming algorithm. Different approaches have been used in the past for accelerating evaluation. For example, it is possible to co-evolve fitness cases in order to reduce the number of evaluations =-=[1]-=-. This, however, adds significant complexity to the algorithm, and does not guarantee an increase in performance under all circumstances. In other applications, one could select the number of fitness ... |
14 | Parallel evolutionary algorithms on graphics processing unit. In: The
- Wong, Wong, et al.
- 2005
(Show Context)
Citation Context ... the most recent information on the projects discussed. Because capable hardware and software are new, there is relatively little previous work on using GPUs for evolutionary computation. For example =-=[11]-=- implements a evolutionary programming algorithm on a GPU, and finds that there is a 5-fold speed increase. Work by [12] expands on this, and evaluates expressions on the GPU. There all the operations... |
13 | in Evolution of Vertex and Pixel Shaders
- Ebner, Reinhardt, et al.
(Show Context)
Citation Context ...find a speed up factor of up to 20. Here both the genetic operators and fitness evaluation are performed on the GPU. Ebner et al, use human interaction to evolve aesthetically pleasing shader programs=-=[14]-=-. Here, linear genetic programming structures are compiled into shader programs. The shader programs were then used to render textures on images, which were selected by a user. However, the technique ... |
13 | Implementing an embedded GPU language by combining translation and generation
- Lejdfors, Ohlsson
- 2006
(Show Context)
Citation Context ...ular choice, and is used for large applications, such as folding@home.sFast Genetic Programming on GPUs 5 PyGPU: Another recent library allows the access of GPU functionality from the Python language =-=[19]-=-. PyGPU runs as an embedded language inside Python. The work is in its early stages, but results are promising. However it currently lacks the optimization required to make full use of the GPU. It req... |
11 |
Visual simulation of shallow-water waves, Simulation Modelling Practice and Theory 13 (8) (2005) 716–726, Programmable Graphics Hardware
- Hagen, Hjelmervik, et al.
(Show Context)
Citation Context ...orithms implemented on GPUs the reader is referred to [7]. For example, discrete wavelet transformations [8], the solution of dense linear systems [9], physics simulations for games, fluid simulators =-=[10]-=-, etc., have been shown to be executed faster on GPUs. In this paper we demonstrate a method for using the GPU as an evaluator for genetic programming expressions, and show that there are considerable... |
9 |
Discrete Wavelet Transform on GPU
- Wang, Wong, et al.
- 2004
(Show Context)
Citation Context ... was cumbersome to use this resource for general purpose computing. For a general survey on algorithms implemented on GPUs the reader is referred to [7]. For example, discrete wavelet transformations =-=[8]-=-, the solution of dense linear systems [9], physics simulations for games, fluid simulators [10], etc., have been shown to be executed faster on GPUs. In this paper we demonstrate a method for using t... |
7 |
Evolutionary computing on consumer-level graphics hardware
- Fok, Wong, et al.
(Show Context)
Citation Context ...ly little previous work on using GPUs for evolutionary computation. For example [11] implements a evolutionary programming algorithm on a GPU, and finds that there is a 5-fold speed increase. Work by =-=[12]-=- expands on this, and evaluates expressions on the GPU. There all the operations are treated as graphics operations, which makes implementation difficult and limits the flexibility of the evaluations.... |
1 | S.M.: Multi-logic-unit processor: A combinational logic circuit evaluation engine for genetic parallel programming
- Lau, Li, et al.
(Show Context)
Citation Context ... possible to offload the evaluation to more suitable hardware. When evaluating digital circuits, they can be loaded into a field programmable gate array (FPGA) and then executed on dedicated hardware =-=[5]-=-. This approach can provide large speed increases. However, the downloading of configurations into an FPGA can be a costly overhead. The biggest drawback to this approach is that it requires the use o... |