Advanced search
Start date

Neuro-GEMA: a grammar-based evolutionary method to automatically design flexible convolutional neural networks


Neural Architecture Search (NAS) is a research field that utilises evolutionary algorithms to generate/optimise neural networks. In recent years, NAS-based algorithms have created models that outperform hand-craft designs in some tasks, settings the new state-of-the-art. In this research project, we will utilise grammar-based multi-objective NAS algorithms to generate deep convolutional neural networks (CNNs). The utilisation of a grammar-based approach to describe and explore the search space allows the researchers to inject domain knowledge into the search process and creates an interface between genotypes and phenotypes that facilitates experimentation, while the multi-objective approach will enable the identification of models with good compromises between predictive power and other metrics, such as model size. The Brazilian partners of this proposal are researchers in the IARA Project (Artificial Intelligence in the Remaking of Urban Environments) funded by FAPESP, a joint nationwide collaboration involving universities and companies in Brazil. Thus, the CNNs discovered are intent to be applied in smart city related problems, such as detecting breeding grounds of disease-carrying mosquitoes and detecting construction materials left on illegal locations, such as sidewalks. (AU)

Articles published in Agência FAPESP Newsletter about the research grant:
Articles published in other media outlets (0 total):
More itemsLess items

Please report errors in scientific publications list using this form.