若手研究者による研究発表2
Tsige Tadesse ALEMAYOH
東北大学 TCPAI 特任助教
Unsupervised Learning of
Time-Series Motion Hierarchies
Traditional AI hierarchies often describe static semantic abstraction, where low-level features form objects and objects form scenes. In contrast, motion time-series are governed by temporal order, variable duration, and the rate of change, making their hierarchy fundamentally different. Hence, time-series hierarchy should be modeled through temporal length and rate of change. This study performs unsupervised learning of a two-level hierarchy of search-and-rescue dog motion, behavior and activity, to reveal behaviors beyond human annotation.
略歴:Tsige Tadesse ALEMAYOH ツィゲ タデッセ アレマヨゥ
Tsige Tadesse ALEMAYOH completed his Master’s and Doctor of Engineering degrees at the Graduate School of Science and Technology, Ehime University, in 2024. Since April 2024, he has been serving as a specially appointed assistant professor at the Tough Cyber Physical AI Research Center. His research focuses on animal motion analysis and the modeling of behavioral policies that enable animals to adapt their motion to the surrounding environment, namely: Animal motion analysis, adaptive behavior modeling.