Costomized algorithm to train a convolutional neural network

Hi, is it possible to train a convolutional neural network, designed with the deep learning toolbox, using a customized algorithm instead of using one of the available 'solverName' in 'options = trainingOptions(solverName)'? E.g. I have designed a convolutional neural network using the toolbox, I need now to train it, but I want to use an algorithm I have designed instead of using SGD with momentum.

Thank you in advance,

Federico

Answer 1

[Shounak Mitra]

This is something that we're working on. We hope to support the flexibility of letting users write custom solvers in the very near future.