- author = "Lorenz Wendlinger and Julian Stier and Michael Granitzer",
- title = "Evofficient: Reproducing a Cartesian Genetic Programming Method",
- booktitle = "EuroGP 2021: Proceedings of the 24th European Conference on Genetic Programming",
- year = "2021",
- editor = "Ting Hu and Nuno Lourenco and Eric Medvet",
- series = "LNCS",
- volume = "12691",
- publisher = "Springer Verlag",
- address = "Virtual Event",
- pages = "162--178",
- month = "7-9 " # apr,
- organisation = "EvoStar, Species",
- keywords = "genetic algorithms, genetic programming, Cartesian Genetic Programming, GPU, ANN, Neural Architecture Search, Reproduction, Object recognition, Convolutional Neural Network",
- isbn13 = "978-3-030-72811-3",
-
DOI = "
doi:10.1007/978-3-030-72812-0_11",
- abstract = "Designing Neural Network Architectures requires expert knowledge and extensive parameter searches. Neural Architecture Search (NAS) aims to change that by automating the design process. It is important that these approaches are reproducible so they can be used in real-life scenarios. In our work, we reproduce a genetic programming approach to designing convolutional neural networks called CGP-CNN. We show that this is difficult and requires many changes to the training scheme, reducing real-life applicability. We achieve a final accuracy of 90.6percent pm0.005, substantially lower than the reported 93.7percent pm0.005. This negates some of the benefits of using CGP-CNN for NAS. We establish a random search as a consensus baseline and show that it produces similar results to the evolutionary method of CGP-CNN. To assess the adaptability and generality of the presented algorithm, it is applied to CIFAR-100 and SVHN with a final accuracy of 63.1percent and 95.6percent, respectively. We extend the investigated NAS by two methods for predicting candidate fitnesses from partial learning curves. This improves CGP-CNN runtime efficiency by a factor of 1.69",
-
notes = "CGP-CNN \cite{Suganuma:2017:GECCO} 1+2 ES Noise due to
GPU implementation??? Random search (only) 2 percent
worse! How to make training CNN faster by 1.7 fold.
NAS=Neural architecture
search.
http://www.evostar.org/2021/eurogp/ Part of \cite{Hu:2021:GP} EuroGP'2021 held in conjunction with EvoCOP2021, EvoMusArt2021 and EvoApplications2021",
