Learning from Non-IID Samples based on the Principles of Lossy Compression

2012 Fiscal Year Final Research Report

• Project/Area Number: 22500119
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Single-year Grants
• Section: 一般
• Research Field: Intelligent informatics
• Research Institution: Gunma University
• Principal Investigator: ANDO Shin 群馬大学, 大学院・工学研究科, 助教 (70401685)
• Project Period (FY): 2010 – 2012
• Keywords: 非 IID データ / 不可逆圧縮原理学習

Research Abstract

In this project, we have developed a principle approach for learning from non-IID data of large-scale information sources. The algorithms developed from this principle were applied to the concrete subjects of physical behaviors of people and autonomous agents. The details of the principles and the algorithms have been published in two international conferences proceedings and with an international journal paper (1). The developed algorithms and benchmark datasets are made public on our website. The algorithms can address the problems of detecting anomalous behaviors and detecting context-specific distributions of behavior patterns, respectively. Furthermore, we developed a general representation model for conducting non-IID data learning presented at oral presentation (2), and a classification model for addressing time-sensitive classification problems presented at oral presentation (1).

Research Products

• [Journal Article] Ensemble Anomaly Detectionfrom Multi-resolution Trajectory Features (2013)
  • Author(s): abc Ando, S.; Thanomphongphan, T.; Seki, Y. & Suzuki, E
  • Journal Title: Data Mining and Knowledge Discovery Pages: 1-45
  • Peer Reviewed
• [Presentation] Time-sensitiveClassification of Behavioral Data (2013)
  • Author(s): Ando, S., Suzuki, E.
  • Organizer: Proceedings of the 13th SIAM International Conference on Data Mining
  • Place of Presentation: Austin,Texas,USA
  • Year and Date: 20130000
• [Presentation] Performance-Optimizing Classification of Time-Series Based on Nearest Neighbor Density Approximation (2012)
  • Author(s): Ando, S.
  • Organizer: IEEE 12th International Conference on Data Mining Workshops (ICDMW)
  • Place of Presentation: Brussels, Belgium
  • Year and Date: 20120000
• [Remarks] 須賀佑太朗,安藤晋,関庸一:人行動分類のための類型パターンに基づく最近傍法.情報処理学会研究報告,2013,2013-MPS-93,pp.1-5