Keynote Speakers

Prof. Daisuke Kitakoshi
National Institute of Technology, Tokyo College, Japan

Title of Keynote Speech
Development of an Intelligent Dialogue Agent as a Component of Comprehensive Preventive Care System

Abstract of Keynote Speech
Demographic aging is recently becoming serious problem in many countries. In particular, Japan, the speaker's mother country is currently known as the most advanced super-aged society while Taiwan will also be one of the top runners soon in terms of demographic aging. This talk focuses on preventive care approaches which aim to decrease the number of people requiring nursing care or support and presents AI-based comprehensive preventive care systems, especially an Intelligent Dialogue Agent (IDA) which plays in important role to mediate between system users (older adults) and the system. Brief introduction of the other components of the preventive care system (fall-prevention part and cognitive training part) is also given in this talk.

Prof. Daniel Hung Kay Chow
The Education University of Hong Kong, Hong Kong

Title of Keynote Speech
RUNNING KINEMATICS MODELING WITH ONE INERTIAL MEASUREMENT UNIT USING DEEP LEARNING

Abstract of Keynote Speech
Kinematic analysis of lower extremity is a popular research approach for running gait analysis. Wearable and mobile sensors would facilitate gait analysis in real running environments. There is an increasing preference for using less sensors or ultimately a single Inertial Measurement Unit (IMU) for such purpose. The current study is to evaluate the performance of a deep learning regression model for predicting bilateral knee and hip flexion during running using only one single IMU worn at the right shank. The target variables of bilateral knee and hip flexion were captured by another IMU based motion capture system for its flexibility to work in different locations. Data from the regressor IMU, together with data from the target variables, were preprocessed and structured as tensor arrays, and fed to a Convolutional Neural Network (CNN) based deep learning model for training. The data were split into different sets for evaluation of different scenarios including: (1) cross-field learning which evaluated how well the model could predict kinematics of different ground surface conditions, (2) cross-leg learning which evaluated how well the model could predict kinematics of the other leg, (3) cross-speed learning which evaluated how well the model could predict kinematic of different running speeds. In all cases, target data for prediction were not present in the learning data and how well the model could predict by virtue of its capacity to generalize to the above unseen scenarios was evaluated. The study also tried to find the best regressor among the variables available in the single IMU which includes three axes of gyroscope and three axes of accelerometer data, and under what circumstance the regressor works best. The results of the experiments will shed light on the possibility of using a single IMU for predicting the lower limb kinematics in a cost effective way.