Institute of Mathematics for Industry

Workshops & Conferences



List of workshops & conferences

Theoretical research to develop new methodology for hypocenter determination based on seismic big data


Hold Date 2017-08-27 00:00~2017-08-30 00:00

Place IMI Conference Room(W1-D-414), West Zone 1, Ito campus, Kyushu University

Theoretical research to develop new methodology for hypocenter determination based on seismic big data

: IMI Workshop of the Joint Research Projects
 
 Date
 
 August 30 (Wed), 2017   ※Closed discussion for other dates.
 
 Place
 
 IMI Conference Room(W1-D-414), West Zone 1, Ito campus, Kyushu University
 Access, Ito Campus map
 
[Program]
 August 30 (Wed)

 10:00 - 10:15    Opening
 Hiromichi Nagao (Earthquake Research Institute / Graduate School of Information Science and Technology, The University of Tokyo)

 10:15 - 10:45
 Speaker: Hiromichi Nagao (Earthquake Research Institute, The University of Tokyo)
 Title: Seismic wave field imaging in the Tokyo Metropolitan area, Japan
 Abstract :
  Long-period ground motions due to large earthquakes can cause devastating disasters, especially in urbanized areas located on sedimentary basins. To assess and mitigate such damage, it is essential to rapidly evaluate seismic hazards for infrastructures, which can be simulated by seismic response analyses that use waveforms at the base of each infrastructure as an input ground motion. The present study reconstructs the seismic wavefield in the Tokyo metropolitan area located on the Kanto sedimentary basin, Japan, from seismograms of the Metropolitan Seismic Observation network (MeSO-net). The obtained wavefield fully explains the observed waveforms in the frequencies less than 0.9 Hz. This is attributed to our seismic wavefield imaging technique, which implements the replica exchange Monte Carlo method to simultaneously estimate model parameters related to the subsurface structure and source information.

 10:45 - 11:15
 Speaker: Shin-ichi Ito (Earthquake Research Institute, The University of Tokyo)
 Title: Four-dimensional variational data assimilation that enables uncertainty quantification
 Abstract :
  The conventional four-dimensional variational (4DVar) method, which is an effective data assimilation (DA) methodology even in the cases of massive simulation models, cannot evaluate, in principle, uncertainties involved in optimum solutions. We propose a new 4DVar-based DA method that enables us to evaluate the uncertainties by implementing a second-order adjoint method. The proposed method is applicable to general autonomous models such as simulations of seismic wave and tsunami.

 11:15 - 12:15
 Speaker: Yoshinobu Kawahara (Osaka University)[Invited]
 Title: Structured Sparse Learning with Submodular Functions
 Abstract :
  A submodular function is a discrete counterpart of a convex function. In this talk, I first review the relationship between structured sparsity and submodularity and then describe how it is useful in several situations in machine learning. In particular, I describe that, in cases with a large subclass of submodular functions that are available as structured regularizers, we can apply efficient maximum-flow algorithms to solve learning problems with the regularizers. Also, I show that Bayesian learning with priors for similar structures is reduced to the similar calculation. I finally show several applications in real-world problems with the developed algorithms.

 12:15 - 14:00    Lunch

 14:00 - 15:00
 Speaker: Yoshiyuki Ninomiya (Institute of Mathematics for Industry, Kyushu University)[Invited]
 Title: Model selection theory in change-point analysis
 Abstract :
  Change-point problems have been studied for a long time not only because they are needed in various fields but also because change-point models contain an irregularity that requires an alternative to conventional asymptotic theory. In this talk, after introducing asymptotic distribution theory about likelihood ratio tests for the number of change-points, we derive Akaike's information criterion for the change-point models.

 15:00 - 15:20    Break

 15:20 - 15:40
 Speaker: Takashi Kurokawa (Graduate School of Information Science and Technology, The University of Tokyo)
 Title: Graph Fourier Principal Component Analysis
 Abstract :
  A principal component analysis (PCA) is a well-known algorithm of data reductions. However, seismic waves at some observatories are correlated and not covered by the algorithm. This speech proposes a new PCA, a graph Fourier PCA applicable for data which is not independent or not identically distributed e.g. seismic observations at some points. The new PCA gives an optimal low-dimensional representation of data in a sense.

 15:40 - 16:00
 Speaker: Tomoya Haba (Graduate School of Information Science and Technology, The University of Tokyo)
 Title: A study of the characteristics of nonparametric Hawkes process
 Abstract :
  Hawkes process is a point process that assumes the history of past events influences the occurrence times of future events. Nonparametric expression of the intensity function enables us to characterize the Hawkes process based on its first- and second-order statistics. We discuss how past events influence the magnitudes of future events in addition to the occurrence times.

 16:00 - 16:20    Break

 16:20 - 16:40
 Speaker: Atori Koie (Graduate School of Mathematics, Kyushu University)
 Title: Multi-class linear discriminant analysis in the case of a large number of classes
 Abstract :
  The canonical discrimination analysis has been widely used for multi-class classification problems. It projects a high dimensional vector onto a low dimensional space by maximizing the ratio of the between-class variance and the within-class variance. In practice, however, the ordinary canonical discrimination analysis cannot perform well when the number of classes is large. To address this issue, we modify the canonical discrimination analysis in two different ways. We apply our proposed methods to the mouse consomic strain data (30 classes).

 16:40 - 17:10
 Speaker: Kei Hirose (Institute of Mathematics for Industry, Kyushu University)
 Title: Sparse covariance estimation via regularization
 Abstract :
  Sparse covariance estimation has been widely used in various fields of research. In this talk, we focus our attention to sparse graphical models, sparse PCA, and sparse factor analysis, and then show some relationships among these models.

 17:10 - 17:30    Closing
 Hiromichi Nagao (Earthquake Research Institute / Graduate School of Information Science and Technology, The University of Tokyo)
 Kei Hirose (Institute of Mathematics for Industry, Kyushu University)

 * Joint Research Period: August 27 (Sun) - August 30 (Wed), 2017.