Machine Learning is alive and well in 2017
It’s fascinating to look at past predictions of the future and compare them to present day reality. In the 19th Century, people believed that miniature Zeppelins would ferry us everywhere, high up in the sky. Arthur C. Clarke’s 2001: A Space Odyssey, presented us with an early vision of machine learning with a sentient computer called the HAL 9000.
HAL was able to think, predict and react to real time events. Unfortunately, poor HAL was conflicted within his initial programming commands and chose to act in a way that went against the wishes of his senior ranking astronaut companions. No spoilers in case you’ve yet to see 2001.
While we don’t have HAL 9000 today, this is a remarkable moment for machine learning. Enabled by ever-improving High Performance Computing (HPC), computers can develop pattern recognition and manage high data volumes to understand an incredible variety of information. Based on data inputs and algorithms, the machine continually learns and makes its own predictions. For example, a computer can learn to spot the difference between a plant and a tree after viewing thousands of images of each.
Machine learning is being put to work in an astonishing variety of highly complex projects, many of which will make a real impact on science and humanity in general. For example, the Research Computing Service at the UK’s Cambridge University is leveraging TOP500 Dell EMC HPC solutions to drive dramatic advances in life sciences, high energy physics and astronomy. They support the Square Kilometer Array project. This multi-radio telescope has a total collecting area of approximately one square kilometre. Fifty times more sensitive than any other radio instrument, an array of this size and power needs machine learning to efficiently manage the massive volumes of data it generates.
According to Paul Calleja, Director of Research Computing at Cambridge, systems with these kinds of performance requirements need to be architected differently from even standard HPC environments. His group has implemented a system aimed capable of large scale High I/O data intensive science research programs. His toolset includes the Wilkes-2 supercomputer (Ranked #100 on the June 2017 TOP500 list). Wilkes-2 is based on PowerEdge C4130 servers. His other machine, Peta4-KNL (#404) is powered by PowerEdge C6320p servers with Intel Xeon Phi (KNL) processors.
Across the Atlantic, MIT’s Lincoln Supercomputing Center (LLSC) has deployed a new HPC system that provides researchers with a dramatic increase in interactive, on-demand HPC and big data capabilities. Powering machine learning, the LLSC’s Dell EMC 648-node HPC system enable advances in research for space observations, robotic vehicles, cyber security, sensor processing, electronic devices, bioinformatics, and air traffic control. LLSC was able to deploy the system through an Intel early access program for the Intel® Xeon Phi™ processor. The Center’s “TX-Green” system is one of the largest of its kind on the US East Coast. It exceeds one petaflop, which quadrupled the LLSC’s computing capacity.
In mainstream IT, Mastercard is realizing tangible business benefits from HPC through machine learning and big data analytics workloads. The company is pioneering the use of machine learning to detect security anomalies or false positives as well as to support human decisions around fraud. Their machine learning solution is also able provide early warnings in the case of “zero-day” malware attacks. At the same time, Mastercard has found it has to bolster its data defenses to safeguard the large volumes of data generated by their HPC infrastructure. It’s an inviting target for malicious actors.
Machine learning is also being put to work in fighting the world’s most deadly diseases. At the Intel® Parallel Computing Center (IPCC) at Dana-Farber Cancer Institute (DFCI), researchers are using a next-generation HPC platform for structural biology based on Intel® Many Integrated Core Architecture (Intel® MIC Architecture). With this powerful tool, IPCC can implement novel machine-learning methods for massively parallel cryo-EM data processing. The result is a unique capability to separate signal from noise and perform 3D reconstruction of molecules that are impossible to identify with traditional technology. Now, they can visualize biochemical molecules in action in ways that have never been done before.
We never stop learning ourselves so we encourage collaborations with academic and commercial organizations around the world to design solutions that meet real, workload-specific requirements. You’re invited to visit the community Machine Learning Knowledge Center, join the Dell EMC HPC user community, and/or visit the Dell EMC TechCenter. To learn more about Dell EMC HPC and machine learning offerings, please visit http://www.dellemc.com/ai