Software-based coprocessor demonstrates relevance for large-scale commercial optimization applications
SAN DIEGO, Aug. 1, 2018 /PRNewswire/ — MemComputing, Inc. today announced it has delivered integer linear programming functionality as part of its proprietary MemCPU™ Coprocessor technology. The company’s software-based coprocessor demonstrated unprecedented computational performance by finding solutions for a problem considered intractable. MemComputing found solutions to other industrial problems orders of magnitude faster than any commercial solver in the market. Through a third party, the company’s technology was evaluated against a well-known set of open, or unsolved, problems from MIPLIB 2010 (Mixed Integer Programming LIBrary) – assembled by academia and industry and maintained by the Zuse Institute of Berlin (http://miplib.zib.de). Since its inception, MIPLIB has become a standard benchmark used to evaluate the performance of mixed integer linear optimization software.
Via the third party, MemComputing addressed one of the hardest problems in MIPLIB which was considered intractable and is frequently part of an annual competition (http://www.satcompetition.org/) – the solution has eluded researchers for almost a decade. Problems of these types may never be solved using currents methods. The MemCPU found a feasible answer within 60 seconds running on a standard computer.
To extend MemComputing’s benchmark achievements, the third party evaluated the MemCPU on a set of problems that represent real-world industrial challenges. MemComputing found solutions at least 10X and 100X faster than the leading commercial solver software.
Fabio L. Traversa, PhD, CTO and co-inventor of MemComputing commented, “We approached these complex problems using an unoptimized prototype of our novel technology – and found feasible solutions for the most challenging problems. We performed significantly better in finding high quality results as compared to the leading commercial solver within the same time threshold. Depending on the complexity of the instance, we found them in seconds or minutes, versus hours or days.”
The MemCPU showed a significant performance gain on the following industrial problems from MIPLIB:
- Efficiency Planning: MemComputing showed linear scaling on a set of problems directed at open pit mining excavation efficiency planning. All other commercial solutions scale exponentially as the problems increase in complexity. With linear scaling, MemCPU enables companies to address problems previously considered impossible in size and scale.
- Network Flow Optimization: Autonomous systems in all industries employ optimization to manage network traffic. For this problem, MemComputing found a high-quality score within five minutes; and running for an hour resulted in scores with even greater precision. Comparatively, a commercial solver would need to run for years and would likely produce inferior results.
- Bioinformatics: Tanglegram Optimization is a common problem in Genomics, Infectious Disease Research and Computational Biology. MemComputing found the best solution previously generated by a leading commercial solver 100X faster, and found even better solutions that are not within reach by current approaches.
“Finding solutions to these problems validates MemComputing’s ability to attack real-world computational problems that have been considered impossible,” said John Beane, CEO of MemComputing. “Businesses that face these problems often require a prohibitive amount of compute time and the growth of Big Data is compounding this issue. Our technology delivers solutions in a fraction of the time driving operational efficiencies that improve a company’s bottom line.”
“MemComputing is actively looking to partner with customers that confront problems of this magnitude,” added Mr. Beane.