The company Mitsubishi Electric Corporation announced the creation of a new high-speed deep machine learning algorithm, which includes functions for inference identification, detection and prediction of the unknown on the basis of the known facts.
According to the developers, their algorithm will facilitate the implementation of machine learning in automobiles, industrial robots and other applications thanks to a substantial reduction in the use of memory and CPU time. In addition, there is the ability to create low-cost solutions for artificial intelligence systems.
Mitsubishi plans to introduce its development during the International Conference on the processing of neural information (ICONIP 2016) which takes place at Kyoto University from 16 to 21 October. In addition, work will be published in the edition of Lecture Notes in Computer Science.
According to the authors of development, algorithm wins for training time, the cost of computing and memory requirements 30 times the typical artificial intelligence systems. Intelligent high-level inferences can be integrated directly into embedded systems without the need to use the server and “heavy” network equipment. Traditional machine learning algorithms for deep require the construction of large neural networks and a large number of resources, including a lot of memory.
According to analysts of the company Ernst & Young, artificial intelligence technology market amounted to about $ 35 billion. In 2016 it will grow another 30% in 2015.