NEC's RAPID machine learning supports human decisions with artificial intelligence, improving work efficiency!
Model size reduced to 1/50 and learning computations reduced to 1/14 of the conventional method
- MitaCan you tell us about the features of RAPID machine learning in a bit more detail?
- TomonagaThey've actually been researching machine learning technology at NEC Laboratories America for over 10 years. This, of course, includes extensive research into deep learning. The core engine of RAPID machine learning is based on technology developed there, so a lot of ingenuity has gone into its calculation methods and algorithms, and the fact it runs fast and is memory efficient is one of its features.
- FujishigeFor example, in tests performed by NEC Laboratories America, we found that RAPID machine learning enabled model size to be reduced to about 1/50 and learning computations to about 1/14 of those for conventional image recognition techniques (leading competitors), without sacrificing accuracy.
- MitaIncredible! That must mean it runs smoothly even on PCs.
- TomonagaThere is another factor that sets NEC apart from other companies.
- MitaWhat would that be?
- TomonagaGenerally, most deep learning software is open source*. There are a number of well-known implementations, including one called Torch. Torch is also used by companies such as Google, Facebook, and Twitter, but it was actually originally developed by NEC Laboratories America. *Software for which the source code that serves as its blueprint is made freely available to the public.
- MitaHuh? Really!? I never knew that open source software developed by NEC was being used at so many other companies.
- TomonagaIf you take a look at the NEC Laboratories America site*, you'll find some information on Torch 5. Rapid machine learning was developed based on this Torch software, and although we've been doing this longer, I believe we have differentiated ourselves from competing companies significantly in the area of innovation as well.
* Information on Torch 5
Larger viewWe have also put a lot of effort into screen operation for RAPID machine learning, and its ease of use is another feature.
Supports advanced decision-making tasks in a range of situations to improve work efficiency!
- MitaCan you give us some concrete examples of how RAPID machine learning can be utilized?
- FujishigeIt can be used to have computers handle or support advanced decision-making that was carried out by humans in the past. Although it can be put to a range of purposes, image analysis and human resource matching are two areas with particularly demand need for it.
- MitaImage analysis and human resource matching? How is it used with these?
- FujishigeFirst, let's discuss image analysis. For example, overseas there is a need to detect people riding motorcycles in pairs from security camera images, to prevent incidents such as bag snatching. Using RAPID machine learning, you can train a computer with scenes where a motorcyclist riding double passes by and scenes without one passing by. That makes it possible to detect when a double riding motorcyclist is shown on a security camera, and notify the supervisor with an alarm, etc.
- TomonagaUntil now, it was necessary to have surveillance staff perform monitoring 24/7, but with computer support they will only need to react when an alarm goes off. This also means monitoring can be carried out by fewer people, perhaps leading to reduced labor costs, and it also enables a wider scope for monitoring.
- MitaThat would make security companies happy!
Larger viewComputers contribute to improved work efficiency and monitoring accuracy for monitoring tasks by handling or supporting anomaly detection
- FujishigeThe technology can also be put to a range of other uses, such as finding people likely to shoplift from store security camera footage, detecting defective products from inspection camera images at factories, and improving the accuracy of medical diagnosis. Incidentally, as another medical example, NEC Laboratories America's "e-Pathologist" technology has also been commercialized by NEC. It is used in the diagnosis of cancer.
- TomonagaUsing image analysis based on RAPID machine learning may make it possible to identify subtler moods from camera images, such as when someone looks like they are enjoying themselves or appears sad. We believe this technology is an extremely good match for things that people find it difficult to identify features for, but can still discern for some reason.
- MitaI guess it could be used to tell whether viewers of a TV program or movie are enjoying it from their facial features! What about human resource matching, then?
- TomonagaHuman resource matching mainly applies to fields such as recruitment agencies and corporate personnel departments. For example, the matching of job seekers looking for mid-career positions and companies who want to hire personnel. When a corporation is hiring new graduates, this technology can also reduce the human workload associated with filtering job seekers a certain extent before hiring.
Larger viewAutomation of the advanced matching of human resources and companies previously performed by specialists. This raises work efficiency and contract rates.
- MitaHow is RAPID machine learning used specifically?
Menu starts here.
End of menu.