IBM today announced the launch of its new Deep Learning as a Service (DLaaS) program for AI developers. With DlaaS, users can train neural networks using popular frameworks such as TensorFlow, PyTorch, and Caffe without buying and maintaining costly hardware.
The service lets data scientists train models using only the resources they need, paying only for GPU time. Each cloud processing unit is set up for ease-of-use and prepared for programming deep learning networks without the need for infrastructure management from users. According to a white paper published by IBM researchers working on the project:
Users can choose from a set of supported deep learning frameworks, a neural network model, training data, and cost constraints and then the service takes care of the rest, providing them an interactive, iterative training experience.
In order to use the service users just have to prepare their data, upload it, begin training, then download the training results. It seems fairly straightforward and could potentially shave days or weeks off of training times.
IBM seems to be working towards taking the difficulty out of training neural networks, or at least lowering the bar for entry. According to a company blog post:
This Deep Learning as a Service is an experiment-centric model training environment, meaning users don’t have to worry about getting bogged down with planning and managing training runs themselves. Instead, the entire training life-cycle is managed automatically and the results can be viewed in real-time and revisited later. Each training run is automatically started, monitored, and stopped upon completion, saving users time and money as they only pay for the resources they use.
A single GPU setup can, for example, take nearly a week to train a visual image processing neural network on a couple million pictures. With IBMs cloud solution that could potentially be cut down to mere hours.
This is further evidence that AI is preparing to infiltrate every industry at just about every level of business. Not only does IBM’s DLaaS give developers access to hardware that’s relatively cost-prohibitive for smaller startups or companies with limited budgets, but it also saves on salary.
Maintaining deep learning systems requires manpower, and this investment of time is exacerbated as projects scale: clustering just a few GPUs for deep learning models is an entirely different skillset than training neural networks.
Rather than require all data scientists to become Jacks or Jills of all trades, a cloud based solution could save money and time. And there’s few companies more equipped to provide DLaaS than IBM.
The company’s DLaaS runs on its excellent Watson platform. This means its been tested with one of the most advanced AI systems on the planet.
For more information you can check out IBM’s blog.