国立大学法人東北大学電気通信研究所公式ウェブサイト

Interdisciplinary Collaboration Research Division Research project of human value estimation of multimodal information based on informatics paradigm to manage both quality

>> Research Center for 21st Century Information Technology(IT-21 Center)

Interdisciplinary Collaboration Research Division Research project of human value estimation of multimodal information based on informatics paradigm to manage both quality

Researcher

  • [ Project Leader, Professor* ] Satoshi Shioiri
  • [ Professor* ] Nobuyuki Sakai
  • [ Assistant Professor* ] Kosuke Yamamoto

Research Activities

There are predictions of data growth beyond one Yotta byte (1024) in 2030, Prioritization of data is critical to deal with such enormous data. Recommendation systems and Curation systems are used frequently but with fixed criteria. These criteria are usually determined by a mechanically along a dimension related certain value estimation. We should realize the system to create a criterion chosen considering the purpose and type of information, and the key technology for the purpose is the one that evaluates quality and value of the information.
To investigate evaluation of a variety of quality and value, knowledge of the brain processing of different types of information is necessary. In addition to vision and audition, which have been major research fields of data evaluation related to human perception, haptic, olfactory and gustatory perception are necessary to investigate in terms of qualities and values. Controlling more senses improves the quality of information communication and makes communication more valuable. To develop the technology for the purpose, investigation of human sense, including haptics, olfaction and gustation, is essential. The present project focuses on multimodal perception related to foods and investigate the quality and value of multimodal food information.

Psychological experiment on interaction between visual information processing with olfactory qualitative and hedonic information processing.
Interdisciplinary2
Fine tuning of a trained network (AlexNet) for our lunchbox data (Top).
Correlation between ground-truth (human judgments) and prediction(Bottom).MAE indicates mean average error.