Abstract:

The objective of this paper is to present an attempt to estimate the training time of convolutional neural networks on single-chip minicomputers. Thanks to this, it will be easy to estimate the time needed to complete the learning process with much slower devices. Using the Keras framework, it was possible to embed identical structures of artificial neural networks without changing the program code from a desktop computer to a minicomputer. In the paper RaspberryPI platform was used to run the tests.