Currently one can choose from pre-defined models (LeNet, AlexNet and GoogleNet). How can I create a custom model (for a non-image or text dataset) and import it into Deep Learning SDK? Will the model have to be written in DAAL or MKL?
The first technology preview release supports only pre-defined LeNet, AlexNet and GoogleNet topologies. We will implement possibility to import custom models in the future releases.
The training tool currently supports only image datasets.
As an alternative, you may connect to docker container of the training tool with pre-installed Intel distribution of Caffe with MKL and run training using console commands.