원문정보
초록
영어
This paper deals with machine learning through interaction with a person. We developed an interactive learning system (LTIP) by concentrating on a specific situation where a user and a robot cooperatively find a lost object in real world. The LTIP consists of a gathering candidate images process, a self-categorizing process based on SOM, and a cooperative and repeatable narrow down process so that the robotic system effectively learns the appearance of the target object that the user is looking for. We conducted several experiments to evaluate and analyze the process of interactive learning. We confirmed that preliminarily determining the effective features set for categorization is difficult. So our proposed LTIP scheme is useful for such a practical processes such as finding a lost object.
목차
1. Introduction
2. Related Works
3. LTIP: Learning Through Interaction with Person
3.1. A Scenario of Finding a Thing
3.2. Bing Search API
3.3. Features of Images
3.4. Clustering of Images by SOM
3.5. Narrow Down Process for Identifying the Object
4. Experimental Evaluation
4.1. Experimental Setup
4.2. Analysis Method
4.3. Experimental Results
5. Conclusion
Acknowledgment
References