A holistic approach for resource-constrained neural network architecture search Lupión, Marcos Calvo Cruz, Nicolás Martínez Ortigosa, Eva Martínez Ortigosa, Pilar Artificial neural networks Neural architecture search Meta-heuristic This work has been funded by the projects ECSEL Joint Undertaking in IMOCO4.E (EU H2020 RIA-101007311) (PCI2021-1219 257) from MCIN/AEI/10.13039/501100011033 and EU NextGenerationEU/PRTR; R+D+i PID2021-123278OB-I00 and PDC2022-133370- I00 from MCI-N/AEI/10.13039/501100 011033 and ERDF funds; and the Department of Informatics of the University of Almería. M. Lupión is a fellowship of the FPU program from the Spanish Ministry of Education (FPU19/02756). N.C. Cruz is supported by the Ministry of Economic Transformation, Industry, Knowledge and Universities from the Andalusian government (PAIDI 2021: POSTDOC_21_00124). The authors would also like to thank J.J. Moreno from the University of Almería for his technical support with the computing platform. The design of Artificial Neural Networks (ANN) is critical for their performance. The research field called Neural Network Search (NAS) investigates automated design strategies. This work proposes a novel NAS stack that stands out in three facets. First, the representation scheme encodes problem-specific ANN as plain vectors of numbers without needing auxiliary conversion models. Second, it is a pioneer in relying on the TLBO meta- heuristic. This optimizer supports large-scale problems and only expects two parameters, contrasting with other meta-heuristics used for NAS. Third, the stack includes a new evaluation predictor that avoids evaluating non- promising architectures. It combines several machine learning methods that train as the optimizer evaluates solutions, which avoids preliminary preparing this component and makes it self-adaptive. The proposal has been tested by using it to build a CIFAR-10 classifier while forcing the architecture to have fewer than 150,000 parameters, assuming that the resulting network must be deployed in a resource-constrained IoT device. The designs found with and without the predictor achieve validation accuracies of 78.68% and 80.65%, respectively. Both outperform a larger model from the recent literature. The predictor slightly constraints the evolution of solutions, but it approximately halves the computational effort. After extending the test to the CIFAR-100 dataset, the proposal achieves a validation accuracy of 65.43% with 478,006 parameters in its fastest configuration, competing with current results in the literature. 2025-12-18T11:56:56Z 2025-12-18T11:56:56Z 2025-03 journal article Lupión, M., Cruz, N.C., Ortigosa, E.M., Ortigosa, P.M. A holistic approach for resource-constrained neural network architecture search. Applied Soft Computing, 172. 112832 (2025). https://doi.org/10.1016/j.asoc.2025.112832 https://hdl.handle.net/10481/108947 10.1016/j.asoc.2025.112832 eng http://creativecommons.org/licenses/by-nc-nd/4.0/ open access Attribution-NonCommercial-NoDerivatives 4.0 Internacional Elsevier