Clients Use Cognitive Computing to Solve Business Challenges
Illustration by Kai and Sunny
Through machine vision and image recognition technology, energy suppliers can train computer systems using visual data to recognize what an intact pipe or high-tension line looks like. They then can fly a drone over a facility or line. With machine speeds of 32 million frames seen per hour, the computer can analyze the drone-generated video and identify if and where to send maintenance personnel.
Healthcare is another data-dense industry where AI technology is used. Radiologists, for instance, look at high-resolution images with subtle differences to determine when clinical issues arise—and whether a diagnosis must be made. To help efficiently read these images, NVIDIA has trained a data set on what an MRI or X-ray might look like across different clinical conditions.
While highly experienced radiologists may take five to 20 minutes to analyze images, NVIDIA also sped up the time it takes for a computer to inspect them. The system provides critical, human levels of analysis but allows radiologists to focus their time on outliers that point to a potentially critical condition.
Pharmaceutical companies use deep learning to analyze high-resolution crystallography images. By using complex image data, they can design new and more effective medications.
These are just a few ways deep learning is being put to work. With PowerAI, more enterprises can take advantage of these data analytic tools.
The Power of PowerAI
In 2015, IBM began to develop PowerAI and integrate it into the Power* platform. “One of the things we found when we embarked on this work last year is that deep learning and the supporting libraries that enable it to work have a lot of complexities and intersections,” Soutter notes.
With PowerAI and the IBM S822LC for HPC, “we have a high-speed interconnect between the different engines inside of the server,” Soutter explains. “This allows the GPUs to communicate rapidly with each other for peer-to-peer computing. And it allows us to move data between system memory in each GPU 2.5x faster than using the competing technology interconnect. This has a dramatic impact on the efficiency of compute time.”
PowerAI is available for download at no charge at ibm.biz/powerai. Any IBM client with a GPU-accelerated compute node can take advantage. The open-source distribution allows PowerAI to be downloaded and installed in minutes. In most cases, it can be up and running in 45 minutes.
“We’ve made this available to all of our clients because we believe it’s one of the most exciting technologies shaping computing today,” Soutter says. “We want to partner with them to make this the foundational technology for deep learning in the enterprise.”
Enterprises must interpret an ever-increasing quantity of digital data in order to make better decisions and create more satisfactory customer experiences. The PowerAI platform for deep learning can help organizations make sense of that complexity—and profit from it.
To learn more about the S822LC for HPC hardware and other scale-out servers, see “IBM Releases Scale-Out Server Guide,” linked in the sidebar.