未学習行動推定のためのZero-shot学習法における精度向上の試み

Moe Matsuki, Sozo Inoue,
人工知能学会全国大会
(Not Available)
(Not Available)
4 pages
2017-05-23
Nagoya
Mobile Activity recognition is important for healthcare and care-life of elderly people. However, there is a problem that the method can’t recognize activities which don’t appear in the training data. We propose the Zero- shot learning with semantic word vectors to overcome the problem. In this paper, we evaluate the proposed method by using 5 supervised learning method to improve the accuracy. As a result, the accuracy of the unseen classes improved than naive methods, and identified several challenges.

Data Files