Abstract | ||
---|---|---|
The machines can work like human or even better with the rapid development of the artificial intelligence (AI) in nowadays. More and more researchers have taken the demand about the interactivity of machine seriously. Human motion recognition that is a fundamental topic in robotics that try to make the machines to understand humans action. While the existing frameworks of motion recognition often require large amounts of training data and cannot extract prior knowledge for next recognition task. In this paper we propose a new framework to make machines have the ability of one-shot learning and learning to learn by using concept learning method. Concept learning decomposes high-lever motions into a series of low-lever motions, fundamental features knowledge was learned from low-lever motions and then the high-lever motions can be learned by just one training data with the help of the fundamental knowledge. The conductive experiments show that the proposed method make machine work efficiently and generally in motion recognition. |
Year | DOI | Venue |
---|---|---|
2017 | 10.1109/I2MTC.2017.7969730 | 2017 IEEE International Instrumentation and Measurement Technology Conference (I2MTC) |
Keywords | Field | DocType |
artificial intelligence,AI,human motion recognition,one-shot learning,concept learning method | Training set,Interactivity,Motion recognition,Computer science,Concept learning,Human motion,Artificial intelligence,Trajectory,Robotics,Machine learning,Learning to learn | Conference |
ISBN | Citations | PageRank |
978-1-5090-3597-7 | 0 | 0.34 |
References | Authors | |
14 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Xiang Ma | 1 | 0 | 0.34 |
Tong Zhao | 2 | 0 | 1.01 |
Ruoshi Wen | 3 | 0 | 1.01 |
Zhaojun Wu | 4 | 17 | 4.27 |
Qiang Wang | 5 | 601 | 84.65 |