Breadcrumb navigation

Message for students 2017: Soma Shiraishi

February 1, 2017

Soma Shiraishi

Creating solutions that can be used by society

Soma Shiraishi
Researcher, Data Science Research Laboratories

Shiraishi has been involved in image recognition since graduate school, and joined NEC Central Research Laboratories in 2013. Since then, he has been developing numerous solutions for society centering on object recognition for retail stores. He is also engaged in practical research that can be applied in the workplace.

Image recognition that unearths the essence of human "understanding"

I currently belong to the object recognition team at the NEC Data Science Research Laboratories, where I am conducting research into image recognition. Image recognition is a research topic I have been engaged in for a long time now, since I entered graduate school, but the reason I chose this topic was simply that I had an interest in the technology itself. Doesn't it sound interesting to study how humans understand things and to try to replicate this using machines? In addition, the development of recognition technology has a direct effect on the lifestyles of people all around the world, in making life more convenient. For me, this is what makes image recognition such an appealing area to research.
My team is researching recognition of objects, but within our group there is also a team that is researching recognition of people. NEC's face recognition technology is one of the best in the world because of this. These days we often talk about how, in the future, we would like to gather together all our team's knowledge and integrate human and object recognition, to create even more advanced recognition technology. Until now, recognition technology has been about identifying a dog from an image of a dog or being able to say who a person is from their face. However, we are entering an era that goes further into the relationships between people and objects. We refer to this among ourselves as "high-level recognition" or "deep recognition." If, for example, it is possible to determine from an image or video "Who is petting the dog," whole new possibilities emerge. This technology is closely related to recent research trends such as artificial intelligence (AI) and deep learning, and it is a topic that I would like to examine more closely in the future.

Retail store solutions to optimize ways of selling

Interview cut of Soma Shiraishi

Since I joined NEC, the main research I have been involved in is the development of solutions for retail stores. For example, one of these was a cash register able to register products by recognizing the products themselves through image recognition. Not only does this make barcodes unnecessary and increase cash register processing efficiency, there is also the benefit that barcode stickers no longer need to be affixed to things like fruit and vegetables. In particular, in supermarkets overseas, fruit is often sold as it is without being packaged, so it is an important benefit to be able to recognize fruit in this way. This technology is still in the development stage, but it is a system that will eventually make unmanned cash registers possible.
In addition to this, I have been involved in developing services that use instore camera images. Images can of course be used to detect mass relocation of products and prevent crime by rapid notification of theft. But we are also focusing on the more positive approach of applying the technology to optimize ways of selling. For example, we can visualize information on product movement, such as if a customer sees a product and takes it down from the shelf to look at it, but then puts it back. This enables stores to examine more efficient ways of selling, such as changing the location of the product on the shelf. The important point is to use the approach of recognizing the movement or changes of products.
At present, we are thinking primarily in terms of use in retail stores, but in the future, we also have in mind the application of this technology to detection or prediction of disasters, detection of suspicious objects in public spaces, and so on. If a machine can recognize what is going on in real time from video images, and can clearly and reliably inform people of the situation, it can help to create a safer and more secure society. I would like to continue with my research toward solutions that can be applied to real life.

Having a sense of balance to combine both researching No.1 technologies and applying them

From my research until now, I really feel the importance of "creating technologies that can actually be used." First, a technology cannot mean anything until it is used to create a service that is needed, and a technology that is developed while sitting at a desk without referencing the real world does not work well in real life. When you actually bring a technology into the workplace to apply it, there are always unanticipated elements and additional innovations that are needed to realize stable operation.
In this sense, too, researchers need to have a sense of balance. Of course, it is important to conduct your research as research for its own sake. All NEC researchers share this common stance of seeking to create a No.1 technology. However, at the same time, it is also important to use the approach of thinking about the actual issue you are trying to solve from a solutions perspective. But neither of these approaches is sufficient alone, and it is not possible to commercialize a product without achieving a balance between the two. I hope that more people with this sense of balance will come and join our team (laughs).
I am currently aiming to become "a researcher who is able to uncover new issues myself." As researchers, the normal way of working is for us to be provided with an issue to solve by a business division, and to work on this issue to find a solution. But I think that if I can identify issues myself and then go on to solve them, I will have become a fully fledged researcher.