Aiming to create a new world inspired by artificial intelligence and sensing technology

Through basic research on machine learning, which is the heart of artificial intelligence, using real-world data from all things connected to the Internet, we predict the future. Furthermore, we generate novels, videos, CG, and animation automatically, in which we are challenging the research of environmental intelligence. Environmental intelligence aims to give intelligence to artifacts, nature, and the environment. For example, “Create your favorite things as intelligent robots,” “Create content in the environment and things around you,” “ We will create a world in which people and things “we create content together.”

Our Research

Interpret the data collected by the sensor

We are trying to make a system that attaches sensors to things around us and to understand what is going on in that environment. For example, an accelerometer is attached to a cup of water to collect data when various people handle the cup of water. By mining them, the relationship between various movements and the data becomes clear, whether you are drinking tea or washing a cup of tea.

Predict abnormal weather

In order to predict the occurrence of abnormal weather that has never been seen in the past, (1) for every 10 years of hourly precipitation data at 1300 locations nationwide, (2) one precipitation pattern in 24 hours Calculate the "abnormality" to determine whether the thought pattern is abnormal. For example, in an area where there is little rainfall for 24 hours in a row, we consider that the rainfall pattern that has been heavily rained for 24 hours has a high degree of abnormality, and (3) a pattern with a high degree of abnormality and a pattern with a low degree of abnormality are repeated. When a pattern that happens or suddenly has a very high degree of abnormality is detected, it is determined that the occurrence of abnormal weather is near.

The position of the smartphone is infered from the sensor value

We can easily obtain the sensor data from a lot of sensors such as a temperature sensor, humidity sensor, and accelerometer equipped in the smartphone. By linking those data with the surrounding environment, you can tell where you are now. This technology is particularly valuable in places where GPS is not available, such as underground shopping malls.

Fusion of two animated images

We are researching to automatically create a face image with the features of two face images, that is, to create a new image by fusing the two images. The method uses deep learning to learn how to digitize image features such as color and outline using a large number of facial images. Then, the features of each image are digitized, and operations such as addition and multiplication can be performed. For example, by averaging each digitized feature, a new feature with each feature of two face images can be obtained.