NEC and AIST develop efficient technology for discovery of rare critical events by merging AI and simulation technologies

New technology reduces the risk of undiscovered critical events and significantly shortens validation time

May 11, 2018 NEC Corporation National Institute of Advanced Industrial Science and Technology (AIST)

NEC Corporation (NEC) and Japan's National Institute of Advanced Industrial Science and Technology (AIST) today announced a new efficient search technology for the discovery of rare critical events. The technology repeats simulations while artificial intelligence (AI) learns the simulation results. As a result, it efficiently discovers rare critical events that are difficult to discover at the product design stage due to the extremely low probability of their occurrence. The technology is one of NEC's lineup of cutting-edge AI technologies "NEC the WISE" (*1).

The new technology merges AI and simulation technologies to improve the efficiency of search for rare events. It significantly shortens validation time (the time taken for highly-skilled experts to discover such rare events at the product design stage) and reduces the risk of multiple critical events being overlooked.

The technology was applied to the design of an optical device to detect stray light (*2), whose probability of occurrence is very low (around 1 in 100 million), but critical because it causes significant performance degradation as a noise signal. While skilled experts require approximately one week, the technology dramatically shortened search time: completing the search in around one day and successfully discovered several critical events without any being overlooked.

With this technology, NEC and AIST will contribute to the further improvement of the quality and reliability of products and operations by supporting human decisions in the design, production, and operation of complex systems, such as social infrastructure; and by discovering and eliminating rare critical events in the design stages in advance.

Technology for the discovery of rare critical events

Features of the rare event discovery technology

• Efficiently narrows down occurrence conditions for rare events and shortens validation time While AI runs simulations, the technology searches for events based on learned characteristics. NEC and AIST have developed an algorithm that enables intentionally biased search based on the frequency of event occurrence. As a result, it has become possible to efficiently narrow down the conditions for rare events, and to discover them in a shorter time.
• Reduces the risk of multiple critical events being overlooked Based on the degree and frequency of occurrence of rare events in its learning outcomes, the AI developed by NEC and AIST calculates and adjusts the concentration ratio of its search in the vicinity of already detected events to 50%, thereby enabling the risk of overlooking multiple critical events to be reduced.

NEC and AIST will extend the field of applications for this technology from optics design to structural design, such as bridges and buildings, and fluid design such as engines, and will continue working in cooperation to contribute to the development of new technologies merging AI and simulations, and to their application in industry.

NEC and AIST presented the technology on May 4 at the SIAM International Conference on Data Mining (SDM18) in San Diego, California.

http://www.siam.org/meetings/sdm18/

• *1 NEC announces new AI technology brand, "NEC the WISE" http://www.nec.com/en/press/201607/global_20160719_01.html
• *2 Stray light: Stray light is unwanted light in an optical system. When stray light reaches the detector, it causes errors and degrades the observational performance of the optical device.