IBM Brings Advanced AI to Data Scientists and Developers to Help Enterprises

March 21, 2018

Artificial intelligence has found its way into hundreds of new applications and industries, and much of that progress is owed to advances in deep learning—a specialty that teaches computers to recognize patterns in data, based on the neurological patterns of the human brain. Deep learning has already helped bring AI into automated translation, medical scanning, self-driving cars and voice recognition. But for all the applications, there are many more waiting to be discovered.

In support of that quest, on Tuesday, March 20, IBM is introducing Deep Learning as a Service within Watson Studio, its integrated end-to-end tooling and training environment. Watson Studio already lowers the barrier to using AI for any application – helping more organizations take advantage of AI – and now IBM is also adding tools to help build deep learning processes, including new drag-and-drop features.

Getting started with deep learning can seem deceptively easy. To start a deep learning process—for instance, teaching a computer to route requests to the proper agent in a call center—all you need is a dataset (such as past phone calls) and an algorithm. All the complexity is beneath the surface, such as layers of parameters and special rules that guide the process. Optimization becomes harder as more layers are added onto the neural network. And most deep learning processes involve many layers.

Most of that complexity can now be handled by Deep Learning as a Service, abstracting away complex tasks for developers. The team has even created a new drag-and-drop interface to speed deep learning application development. Any developer with a basic understanding of machine learning can quickly and efficiently set up new processes. To teach a call center AI, for instance, you’ll be able to connect your phone call dataset, parameters and learning algorithm.

Hyper-parameters are also abstracted: developers, who previously had to intensively study their subject area to choose their hyper-parameters, can now tap Watson to automatically handle this complex process. Deep Learning as a Service will also work with any other open source AI software or popular frameworks like TensorFlow, PyTorch and Caffe.

Lastly, Deep Learning as a Service is a cloud-based solution with scaled pricing. Developers only pay for time spent on the deep learning process. When it’s complete, the process ends. End result: Deep learning doesn’t just become more accessible —it’s on demand.