- author = "Leonardo Trujillo and Jose Manuel {Munoz Contreras} and Daniel E. Hernandez and Mauro Castelli and Juan J. Tapia",
- title = "{GSGP-CUDA} - A {CUDA} framework for Geometric Semantic Genetic Programming",
- journal = "SoftwareX",
- year = "2022",
- volume = "18",
- pages = "101085",
- month = jun,
- keywords = "genetic algorithms, genetic programming, Geometric Semantic Genetic Programming, separate generations, elitism, stack-based, GSGP-C++, parallel computing, SIMD, CUDA, GPU, GPGPU, C/C++/CUDA, CUBLAS",
- ISSN = "2352-7110",
-
URL = "
https://www.softxjournal.com/article/S2352-7110(22)00060-7/pdf",
-
URL = "https://www.sciencedirect.com/science/article/pii/S2352711022000607",
-
DOI = "
doi:10.1016/j.softx.2022.101085",
-
code_url = "https://github.com/ElsevierSoftwareX/SOFTX_2020_38",
- size = "7 pages",
- abstract = "Geometric Semantic Genetic Programming (GSGP) is a state-of-the-art machine learning method based on evolutionary computation. GSGP performs search operations directly at the level of program semantics, which can be done more efficiently than operating at the syntax level like most GP systems. Efficient implementations of GSGP in C++ exploit this fact, but not to its full potential. This paper presents GSGP-CUDA, the first CUDA implementation of GSGP and the most efficient, which exploits the intrinsic parallelism of GSGP using GPUs. Results show speedups greater than 1000 fold relative to the state-of-the-art sequential implementation, during the model training process. Additionally, our implementation allows the user to seamlessly make inferences over new data through the best evolved model, opening the possibility of using GSGP on Big Data problems.",
-
notes = "See \cite{DBLP:journals/corr/abs-2106-04034}. Also
known as \cite{TRUJILLO2022101085}
p2 'a linear representation is used'. 'real-valued symbolic regression.' 'mutation is usually sufficient'. 'compute the semantics and fitness of an offspring using only the semantic vectors.' using 'building and updating [pop_size by number_fitness_cases] semantic matrices'
'maximum program size k.' 'fitness given by the Root Mean Squared Error (RMSE).'
cudaOccupancyMaxPotentialBlockSize
p3 '+, −, * and protected division. ... probabilities: 0.8 for a function, 0.14 for a problem variable or feature, and 0.04 for an ephemeral random constant.' (ERC)
p4 Quadro P4000 (1792 cores), Tesla P100 (3584 cores). 'virtual machine' 64-bit Ubuntu Linux. popsize = 10240, 50 generations.
",
