Fujitsu has developed an AI solution that can recognize universal features that signify concentration or non-concentration without being affected by the subjects' cultural contexts.

FREMONT, CA: Fujitsu Laboratories, one of the premier research centers globally, has revealed the successful development of a new general-purpose AI model for estimating concentration levels that can accurately measure and quantify a person's level of concentration when performing different activities. The model does this by identifying small variations in muscle movements that reveal whether or not a person is concentrating.

Models that use AI to measure attention have traditionally been developed by teaching algorithms to understand the movements and actions of people doing particular activities, such as e-learning. However, since facial expressions and actions vary based on each person's tasks and the cultural context, the models developed had to be unique, and the challenge was to build individual AI models for various, particular circumstances.

Fujitsu has developed an AI solution that can recognize universal features that signify concentration or non-concentration without being affected by the subjects' cultural contexts. With the highest precision in the world, the AI uses proprietary technology to detect Action Units (AU), which express the "units" of activity corresponding to each muscle or muscle group of the face based on an anatomically based classification scheme. Changes over a few seconds, such as a tense mouth, and long-term changes over tens of seconds, such as staring intently, are captured by the technology, with time intervals tailored for each action unit.

To construct a machine learning data set, data was obtained from 650 people from several regions, including the United States, China, and Japan. They were engaged in tasks such as memorization and searching that required attention. This data was used to develop a general-purpose AI model that can detect concentration levels without relying on task-specific behaviors. Using this data collection, the model's precision was confirmed, and it was confirmed that subjects' degree of concentration could be quantitatively measured with an accuracy rate of over 85 percent. Finally, as more and more facets of life shift online in the face of the COVID-19 pandemic, this technology provides AI assistance that allows users to use reliable data about focus and increase their online activities' quality and effectiveness.