Speakers

Quickly Analyzing Sensitivity of Incremental Data Update for Big Data Machine Learning in Changing Environment

I_02_Ichiro takeuchi(Photo)
Ichiro Takeuchi
Nagoya Institute of Technology
15:50~16:30

Abstract :

In this talk, we are concerned with large-scale machine learning problems in changing environment where a small part of the dataset which have been used for training a machine learning model is incrementally updated, and the effect of the data update must be quickly incorporated into the machine learning model. When the entire dataset is large, even if the amount of the data update is fairly small, the computational cost of re-training the machine learning model would be prohibitively large. In this talk, we introduce a novel method called Quick Sensitivity Analysis (QSA) for efficiently incorporating such a data update effect into the machine learning model without spending large computational cost for re-training it. The proposed QSA method provides bounds on the unknown yet-to-be trained machine learning model with the cost only proportional to the size of the data update. We demonstrate through numerical experiments that the proposed QSA method provides sufficiently tight bounds with negligible computational costs especially when a small part of the dataset is updated in a large-scale machine learning problem.

 

Education

  • 2000 Ph.d (Doctor of Engineering), Graduate School of Engineering, Nagoya University
  • 1998 Master of Engineering, Graduate School of Engineering, Nagoya University
  • 1996 Bachelor of Engineering, Graduate School of Engineering, Nagoya University

 

Professional Activities

  • 2015-present: Full Professor, Department of Engineering, Nagoya Institute of Technology,
    Japan (tenured)
  • 2008-2015: Associate Professor, Department of Engineering, Nagoya Instutute of Technology, Japan (tenured)
  • 2001-2008: Assistant Professor, Department of Engineering, Mie University, Japan (tenured)
  • 2000-2001: Post-Doctoral Visitor, Department of Computer Science and Operations Re-
    search, University of Montreal, Canada
  • 2000-2001: Research Fellow, Japan Society for the Promotion of Science, Japan
  • 1999-1999: Junior Research Associate, The Institute of Physical and Chemical Research
    (RIKEN), Japan