EvoDeep: a new Evolutionary approach for automatic Deep Neural Networks parametrisation

We have presented EvoDeep, a tool for evolving Deep Neural Networks parameters and architecture. The approach pursues to maximise its classification accuracy while achieving valid sequence of layers by employing a state machine.

This model is tested against a widely used dataset of handwritten digits images. The experiments performed using this dataset show that the Evolutionary Algorithm is able to select the parameters and the DNN architecture appropriately, achieving a 98.93% accuracy in the best run.

Download full paper here

Please, if you use this paper, cite it!

Martín, Alejandro; Lara-Cabrera, Raúl; Fuentes-Hurtado, Félix; Naranjo, Valery; Camacho, David. EvoDeep: a new Evolutionary approach for automatic Deep Neural Networks parametrisationJournal of Parallel and Distributed Computing, 2017.

Leave a Reply

Your email address will not be published. Required fields are marked *