IBM’s Watson Studio is the company’s service for building machine learning workflows and training models, is getting a new addition today with the launch of Deep Learning as a Service (DLaaS). The general idea here, which is similar to that of competing services, is to enabled a wider range of businesses to make user of recent advances in machine learning by lowering the barrier of entry.
With these new tools, developers can develop their models with the same open source frameworks they are likely already using (think TensorFlow, Caffe, PyTorch, Keras etc.). Indeed, IBM’s new service essentially offers these tools as cloud-native services and developers can use a standard Rest API to train their models with the resources they want — or within the budget they have. For this service, which offers both a command-line interface, Python library or interactive user interface, that means developers get the option to choose between different Nvidia GPUs, for example.
The idea of a managed environment for deep learning isn’t necessarily new, With the Azure ML Studio, Microsoft offers a highly graphical experience for building ML models, too, after all. IBM argues that its service offers a number of distinct advantages, though. Among other things, the service offers a drag-and-drop neural network builder that allows even non-programmers to configure and design their neural networks.
In addition, IBM’s tools will also automatically tune hyperparameters for its users. That’s traditionally a rather time-consuming processes when done by hand and something that sits somewhere between art and science.