Deep Learning, Modeled after the Brain’s Mechanisms
It was once believed that computers "demonstrate their capabilities only after being programmed (or given knowledge) by humans." In recent years, however, studies on artificial intelligence (AI), in which the computer learns by itself and becomes more intelligent just like the human brain does, have progressed and been applied to a wide range of technologies and systems.
Although the word AI may conjure up images of chess or shogi (Japanese chess) matches against professionals, or voice recognition and response technology in smartphones and car navigation systems, the current focus is on machine learning. In machine learning, computers identify and learn consistent patterns or rules found in sets of Big Data. One of the methods for such learning, called deep learning, is currently attracting a lot of attention.
This new technology uses a neural network modeled after the human brain structure. A human brain contains neurons and synapses that link neurons together and pass information between them. Deep learning is the latest technology using this neural network structure and has been applied to the field of image and voice recognition.
Fujitsu Limited announced the systemization of its AI technologies into Human Centric AI Zinrai in November 2015. The company has been studying machine learning. One example of this is the world's first handwritten Chinese recognition technology using deep learning developed by Fujitsu Laboratories Ltd. Reports on this research stated that this technology achieved a 96.7% recognition accuracy rate, which was higher than the accuracy of human recognition. This technology can automate computer input and checking tasks that are traditionally done by humans.
Deep Learning "Understands" the Five Senses
What is the use of deep learning? Deep learning is a technology that can "understand" the five human senses. For example, it may assess freshness of food based on its appearance, issue warnings that the food has gone bad, and prevent food poisoning. Or, if used in a car navigation system, it can alert the driver to predicted danger at an intersection with blind corners.
In addition to such visual usage, the technology can have audio usage. Deep learning about the voice and vocal qualities of an artist may result in a system that recommends songs of other artists who have similar voices.
Deep learning can help training; it facilitates learning for people to whom a job will be handed over by identifying the characteristics of a craftsperson's physical movements so that successors can learn to imitate them.
Computers can find rules that are equivalent to those drawn up by professionals from years of experience.
With the advancement made in realizing the Internet of Things (IoT), a huge volume of data is being collected and stored from a wide variety of devices. Deep learning technology can sort and analyze such Big Data at a high-level of accuracy. This capability is expected to lead to the creation of new value and business fields.
Data Curation Services Support Business Creation
Fujitsu Limited started its data curation service in 2012. Then, it strengthened the services and launched the deep learning-based Big Data analysis service. Curators specialized in data analysis at Fujitsu Limited analyze customers' image and voice data and examine the effectiveness of using deep learning. Customers can use the results of these examinations to consider future business creation or business innovation.
In a conventional data curation service, device logs, customer information, and product information were analyzed and output as a prediction model. The new service utilizes “five-sense” data to support service development and business innovation that match the needs with higher accuracy.
A case example of this new service is the joint project with COOSY Inc., a company that manages one of Japan’s largest cosmetics and beauty information websites, called Hapicana. Fujitsu and COOSY started to develop a new service using deep learning in January 2016. In this project, deep learning technology learned about 50,000 pieces of facial image data. It then created eight facial models. Using these facial models, the new service will recommend makeup methods or items that match the characteristics of users who upload their face photos onto the website.
Future with Deep Learning in Various Business Fields
Fujitsu Limited plans to apply machine image learning using its deep learning technology to the fields of manufacturing, medical care, advertising, and sports. At factories, quality inspections using product images can enhance how accurately finished products reflect the specifications. At hospitals, latent focuses of disease can be found in X-ray images, so resulting in early detection. In trains, advertisements can be changed in accordance with the number of passengers or the direction of their faces. In sports, the characteristics shared by top athletes can be identified and used in instructions.
Fujitsu Limited is promoting its convergence service. In this service, the company creates new value by effectively using Big Data to add more value to customers’ services and products and to help them develop their services and products faster. Fujitsu Limited will actively use the latest technologies, including deep learning, and apply ICT to tasks that have been difficult to automate in order to help customers with their business innovation and creation.