A new algorithm developed by Mitsubishi Electric could simplify the implementation of deep learning in vehicles, industrial robots, and other machinery.
The new system adapts to the specific purposes of each system, using learning data and high-level inferences about the operating environment. This supports the effective structuring of networks and reduces the trial-and-error of design, the company contends.
The algorithm is expected to reduce training time, computational costs, and memory requirements to about 1/30th of that of conventional artificial intelligence. The company’s current compact AI has already reduced the computational costs and memory requirements for image recognition by 90 percent compared to conventional AI.
The system should help the company expand its AI range of utilization thanks to its compact size and low cost.
“It will reduce the costs of AI deployment by eliminating needs for servers and network facilities because of its compactness and high-level inference to be performed directly in embedded systems,” said a Mitsubishi statement. “Conventional machine-learning algorithms for deep learning requirement deep neural networks comprising costly memory resources.”
Mitsubishi will present the new algorithm at the International Conference on Neural Information this week at Kyoto University.
Filed Under: Industrial automation
