Deep learning

Compact algorithm.
Implement high level AI for all equipment.

What is deep learning?

AI, or artificial intelligence, is a technology that uses computers to perform intellectual functions like logical inference or learning from experience, just as humans do.
AI has evolved rapidly in recent years as computing devices have reached higher levels of performance. Nowadays, AI is an important technology supporting our society. Machine learning is one field of AI, and deep learning is one type of machine learning.

Deep learning is based on neural networks, which reproduce the network of human brain neurons as a mathematical model. A neural network is composed of three kinds of layers; the input layer, the hidden layer, the output layer.
By processing information in multiple layers, neural networks are capable of high-level recognition, identification, analysis, etc. There is a great expectation that this technology will make computers more like humans.

Strengths of Mitsubishi Electric

Dramatically less calculation for the same inference accuracy.

There are several issues that need to be addressed for deep learning to become more widespread.
One such issue is the great amount of calculation. It can be a challenge to equip factory automation, automobiles, and other equipment with deep learning because it is so hard to put high-performance computing devices and high-capacity memories on small devices.
Mitsubishi Electric has developed a proprietary algorithm that greatly reduces the amount of calculation while maintaining a high level of inference accuracy.

The input, hidden, and output layers of a neural network connect to each other in complex ways, like tree branches spreading out. A massive amount of calculation is required to process data this way.
Drawing on our machinery knowledge built up over many years, we successfully compacted the amount of calculation to just 1/30 to 1/100 the original amount by “cutting the branches” that are less essential.
This makes it possible to implement deep learning in a wide range of devices and further expands the potential of AI.