Encoding bird's trajectory using Recurrent Neural Networks

Ilya S. Ardakani, Koichi Hashimoto

研究成果: Conference contribution

4 被引用数 (Scopus)

抄録

Recurrent Neural Networks (RNNs) are currently state of art tools for processing and classifying data sequences. This work aims to exploit these capabilities in Long-Short Term Memory (LSTM) networks which are a powerful variant of RNNs for encoding the birds' trajectory data into state vectors. These vectors should encapsulate the contextual information about the immediate trajectory coordinates. Therefore, they can generate new trajectory points based on their state and the latest output. However, probabilistic behavior of birds, effects of environment and noisy nature of measurements pose challenges for training and testing of the LSTM network models. This study solely focuses on the effects of spatial context and their significance in subsequent outputs to achieve compact representation of the traversed trajectory. Therefore, trajectory coordinates of birds were used as input to LSTM networks to learn spatial path features encoded in hidden vectors of the network. In the end, t-SNE method is used to visualize the state vectors in lower dimensional space embeddings and It was observed that these vectors contained contextual information about the traversed path.

本文言語English
ホスト出版物のタイトル2017 IEEE International Conference on Mechatronics and Automation, ICMA 2017
出版社Institute of Electrical and Electronics Engineers Inc.
ページ1644-1649
ページ数6
ISBN(電子版)9781509067572
DOI
出版ステータスPublished - 2017 8月 23
イベント14th IEEE International Conference on Mechatronics and Automation, ICMA 2017 - Takamatsu, Japan
継続期間: 2017 8月 62017 8月 9

出版物シリーズ

名前2017 IEEE International Conference on Mechatronics and Automation, ICMA 2017

Other

Other14th IEEE International Conference on Mechatronics and Automation, ICMA 2017
国/地域Japan
CityTakamatsu
Period17/8/617/8/9

ASJC Scopus subject areas

  • 制御と最適化
  • 器械工学
  • 人工知能
  • 産業および生産工学
  • 機械工学

フィンガープリント

「Encoding bird's trajectory using Recurrent Neural Networks」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル