Does the introduction of the Deep Learning service mean this is the first-time IBM has ever had a deep learning offering (or used deep learning in its Watson services)?

We’ve offered an on-premise version of Deep Learning for quite some time,  but this is the first time IBM offers a deep learning service on the cloud. Plus IBM continues to provide deep learning-based services to application developers with the Watson Developer Cloud.

Why is this deep learning service offered within the Watson Machine Learning Service rather than standalone?

Deep learning is subset of machine learning, so it makes more sense for Deep Learning to be a service within Watson Machine Learning. In addition, users of the deep learning service get the powerful backing of Watson Machine Learning plus easy integration with other services under the Watson Studio umbrella.

What is the Watson Studio relationship to the deep learning service?

IBM delivers deep learning through IBM Watson Machine Learning service which is integrated into IBM Watson Studio.

How does the Neural Network Modeler work with the IBM Watson Machine Learning service?

The Neural Network Modeler works within the deep learning service. Data scientists, developers and business users can design their neural models through a drag-and-drop process without code. The Neural Network Modeler generates the code from one of the user’s preferred frameworks such as TensorFlow, Keras, PyTorch or Caffe.

Is IBM the first/only organization to offer a Neural Network Modeler capability?

No, but IBM provides it’s network modeling in the context of a complete machine learning platform. Models designed using Neural Network Modeler can buildt using IBM’s experiment-centric deep learning service then deploy them as REST endpoints. Neural Network Modeler supports numerous open source frameworks and lets the user choose which they’d like to work in.

Are there any customers using the deep learning service?

The service was previously in a closed beta.

How was IBM Research involved in this process?

The core capabilities of the new deep learning service originated from various projects within IBM Research. More specifically, the following features were implemented directly from IBM Research:

  • The micro-services supporting the deep learning service manages the distributed training of models in parallel across a cluster of GPUs while supporting multiple open source frameworks like Tensorflow, Caffe, Keras and PyTorch.
  • Neural Network Modeler allows for the rapid design of complex networks without coding. The research code name for this project is Darviz.
  • Hyperparameter optimization (HPO) allows the deep learning service to tune parameters of Neural Networks automatically. With this technology we are able to automate the iterations of the hyperparameters to find the best neural network for each use case.
  • Distributed deep learning with Uber’s Horovod and DDL (Distributed Deep Learning).
  • The graphical UX plus model performance tracking powering Experiment Assistant originated with IBM Research with the codename of Project Runway.

Is Watson Studio only available on the Cloud?

Yes.

Does the on-premise version of Data Science Experience offer the same capabilities as Watson Studio?

No, but to learn more please visit Data Science Experience Local.

Is the deep learning service only available on the Cloud?

Yes, it is available as deep learning as a service within IBM Watson Machine Learning service.

How does IBM’s deep learning feature differ from what competitors offer?

IBM’s deep learning offering differs from competitors in numerous ways:

  • Neural Network Modeler (described in more detail above) is not currently offering by our competitors.
  • Experiment Assistant delivers an experiment-centric managed container-based training flow that supports easy monitoring of parallel training runs built using the most popular deep learning frameworks.

What does the pricing plan for this feature look like?