H20 Driverless AI Accelerates Time to Insight for Enterprises
IBMer Fred Robinson explains how the prebuilt AI models in H20 Driverless AI can get clients up and running faster.
By Neil Tardy06/01/2020
Long ago, successful enterprises recognized that their data was their most valuable resource and understood that critical insights could be gained by analyzing these troves of information.
To accomplish this, database developers initially relied upon flat files and query tools. Soon, the options transformed into data warehousing and data mining. More recent innovations allow analysis to extend far beyond traditional relational data to include information found in email, social media feeds and smart devices. And with the advent of AI/machine learning and deep learning, the transformation of analytics only continues to accelerate.
However, the results-oriented world of business approaches the topic with an abundance of caution. It’s something that Fred Robinson, high performance computing (HPC) and AI lead with IBM Systems Lab Services, routinely encounters when helping with client projects.
“Every client that tries a new technology is skeptical,” he says. “Things may look good on a website, but the proof is in the pudding when you try to implement it.”
Lately though, Robinson is seeing skeptics won over by a solution known as H20 Driverless AI on IBM Power SystemsTM. This hybrid offering has a few layers, so let’s jump in and work our way through.
H20 is the flagship open-source solution from H20.ai, the Silicon Valley startup and provider of analytic solutions based on data science and machine learning. In 2018, IBM partnered with H20.ai to create an enterprise version of the machine learning software. The product can be implemented with the assistance of IBM Systems Lab Services or IBM Business Partners.
“The solution allows you to leverage your organic knowledge of your business, data and your relationships with customers by drawing new relationships between various types of data that you may not have been able to previously analyze.”
POWER ServersSupport Driverless AI
Robinson likes to point out that AIX® and IBM i installations know their data. And through every iteration—data warehousing, business intelligence, big data, however you choose to label it—they’ve seen the results and reaped the benefits. Using this information and by developing their customer relationships, they’ve built their businesses. Simply, Driverless AI represents the next and most potent generation of these tools. Driverless AI, says Robinson, is the, “on ramp to the AI/machine learning world, but you don’t have to be a data scientist to obtain value. You just need to be able to understand your data.”
With Power Systems providing the foundation to support the intense data processing and memory requirements of AI workloads, Driverless AI opens up every system in your environment to data visualization, feature engineering and model interoperability. This includes other servers and databases, as well as cloud storage services like Amazon S3, Google BigQuery and Azure Blog Storage. Moreover, with the automation built into driverless technology, it's unnecessary to conduct extensive feature engineering upfront.
Assuming, naturally, that you have sufficient rack space and power, Driverless AI can function in AIX or IBM i environments on any POWER9™ model or POWER8® model, according to Robinson, by downloading Red Hat® or another Linux® distribution. And of course, it also takes advantage of the GPU capability available for Linux on Power® environments.
That said, Driverless AI is optimized for one server in particular: the Power System AC922. Already renowned as the backbone of the U.S. Department of Energy’s Summit and Sierra supercomputers, the AC922 is available in configurations up to 44 cores. Utilizing POWER9 CPUs and NVLink GPUs (powered by the NVIDIA Tesla V100 accelerator), the AC922 provides, according to IBM, up to a 5X boost in performance. Modern GPUs are designed to perform parallel operations on multiple data sets, making them particularly well-suited for AI environments.
Robinson adds that other tools can be easily integrated to support driverless AI. This includes open-source machine learning frameworks like TensorFlow or PyTorch Anaconda, or IBM offerings from its Watson® machine learning or Visual Insights base. Previously called PowerAI Vision, Visual Insights is a deep learning platform that allows video and visual information to be deployed without coding or deep learning expertise.
“With the fast transfer between memory and the GPUs, combined with Power and the capability we have with the I/O bus, everything is more efficient,” says Robinson. “That means faster training time, faster ingest time and faster cycle time throughout.”
Robinson says Driverless AI has had good adoption, particularly among those who are acquiring the AC922 specifically to host the solution. “Working with customers on the initial proof of concept and proof of delivery, they’re finding value, and that’s good to see.”
Of course, in many circles, that familiar skepticism still lingers. “I think a majority of business clients are waiting. It’s like, I’ve implemented data warehousing, I’ve implemented some other tools. I’ve got what I need for now,” Robinson says.
To these individuals, he suggests spending some time on the H20.ai site reading the customer case studies, which recount client experiences with Driverless AI. They’re sorted by industry (financial services, insurance, healthcare, etc.), and Robinson recommends reading up on businesses that are similar to yours to see if something resonates.
“The marriage of H20 Driverless AI and IBM Power Systems is a strong one, and it continues to grow,” he adds. “The solution allows you to leverage your organic knowledge of your business, data and your relationships with customers by drawing new relationships between various types of data that you may not have been able to previously analyze.”
Getting Started with Driverless AI
H20 tutorials: Learn distributed machine learning and chat with data scientists on Slack
Learn Python: Free online courses for the open-source language, along with general data science and AI concepts and programming techniques
IDC Report: An in-depth look at AI adoption, sponsored by IBM and H20.ai
Neil Tardy is a contributing writer to IBM Systems Magazine.