By Armand Ruiz, Stevan Slusher, Adam Massachi, Greg Filla, Julianna R. Delua, Yin Chen and Vishnu Alavur Kannan
At the intersection of data, mathematics, and business you’ll find the most exciting area of technology today. With a background in business, I began to consider the impact data science could have in real-world applications very early on in my career. Over the years since the open source community and amazing evolution of tools and techniques developed around data science and machine learning (ML) have pushed these technologies to maturity. They now positively impact our businesses, our lives, and the world.
Today, I’m excited to announce another step in that evolution of data science with IBM® Watson® Machine Learning (WML) V2.0. WML is a convergence of divergent technologies that enables data scientists and citizen data scientists to more easily and efficiently deploy self-learning models into production at scale.
Now, let’s take a look at the foundational benefits offered by WML, the enhancements afforded by V2.0 and the new tools that data scientists, DevOps and application developers will be using on their continued journey to AI.
Deploy and manage models at scale
Over the course of its history, WML has been helping data scientists and enterprises with the deployment of their analytical assets and models. The solution makes it easier for data science teams and application developers to focus more on the science of getting a highly accurate model created and then making it easy for them to embed it in a system or application.
WML also allows you to perform critical tasks beyond the deploy phase. Once deployed, your models need to be well-managed. For that result, we included versioning controls and the ability to build automation pipelines around models, preventing them from going stale or losing the quality of accuracy in performance over time.
Enable intelligent model operations
With WML, deployed models are continuously learning and updating over time. That’s because embedded intelligent training services feature feedback loops that constantly learn from new data, regardless of where it resides.
Also, WML will retrain and redeploy models as needed and based on how they’re performing in the system or application. This prevents lengthy stretches where your prediction accuracy is degrading; users are notified of any such degradation more quickly. As a result, the models will be retrained more reliably due to automation.
Accelerate compute-intensive workloads
The third overarching benefit of WML comes by way of an enhanced integration being released with WML 2.0, the WML Accelerator add-on. WML Accelerator allows you to push batch training jobs from Watson™ Studio Local to a compute optimized cluster that will accelerate training jobs and provide optimal management of resources. These trained models are available in your WML environment where they can be managed and deployed.
Once a job is in the WML Accelerator environment, WML Accelerator will schedule it and any other jobs currently in the pipeline in the most efficient fashion, allowing all batch training jobs in the current pipeline to be completed quicker than ever before.
This process can be managed and monitored via an easy-to-use built-in user interface (UI), that extends to most areas of WML. In addition, with dynamic resource allocation, we’re making sure that jobs with higher priorities are having the most resources allocated to them.
Bring it all together seamlessly across multiple clouds
Along with Watson Studio, Watson Machine Learning 2.0 brings together public and private cloud technologies to create a unified experience and increased efficiency in the training and deployment process, and it’s all easily managed through a collaborative UI offering the same experience for all vested parties.
Want to train a model on your on-premises environment and deploy to a hosted cloud? You can easily do so with WML 2.0 thanks to its hybrid nature.
WML 2.0 utilizes a common API between cloud and local, meaning that not only can you work seamlessly across environments, you’ll also enjoy programmatical interaction whether using the IBM Watson Machine Learning API, Command Line Interface, or Python Client.
Innovations that simplify the use of models for all
WML V2.0 is a strong move forward in the analytical AI deployment space. And, what’s more, IBM will be providing continued support for the tool with even more exciting features to come, such as AutoAI, coming soon to cloud and local, as well as improved model operations and DevOps.
Want to know more? View the supporting blogs of my colleagues below. Vishnu Kannan covers the capabilities of Watson Studio V2.0 and how the technology works cohesively with WML. Julianna Delua covers the overarching benefits of the combined technologies and how they benefit modern enterprise. Or, you can read the Enterprise Strategy Group (ESG) technical validation for even more detail.