You are welcome to attend the following seminar.

Mr Hoareau (SOKENDAI, NII) and Dr Duval (NII) will present their research on location-awareness in ubiquitous computing and human needs in wearable computing at the National Institute of Informatics.

Schedule: Thursday 04 October 2007 (15:30-17:30)
Location: National Institute of Informatics (Tokyo), 20F, Room 2005
Language: English
Registration fees: None
Seminar page: http://horizons.free.fr/his/eng/seminars.htm

Location-awareness in Ubiquitous Computing (15:30-16:30)
by Mr HOAREAU Christian

Abstract: Over the last decade, devices related to ubiquitous computing (e.g. sensors, embedded microprocessors, handhelds, wearable computers) have soared and seamlessly integrated into our everyday lives.
They provide new functionality, enhance user productivity, and ease everyday tasks. First, I will give an overview of context-aware
services.
Then I will focus on my current research: location-aware services that (1) enable emerging applications to detect the presence/location of people and objects, then (2) behave accordingly.
To support such functions, ubiquitous systems must capture, maintain and process location information.
I will thus discuss location modelling and present how I use location models to build a location query framework.

Speaker: Mr Hoareau obtained a Master's degree in computer science from the Université Pierre & Marie Curie (France).
Since 2006, he is a Ph.D. candidate at 国立情報学研究所 (National Institute of Informatics), 総合研究大学院大学 (The Graduate University for Advanced Studies).
He is currently studying location modelling and query processing for ubiquitous computing, under the guidance of 佐藤 一郎 (SATOH Ichiro).

Needs in Wearable Computing: a User-centred Approach
(16:30-17:30)
by Dr DUVAL Sébastien

Abstract: Technological and algorithmic advances allow the creation of functional wearables, small computers that can be held (e.g. cellular phones) or worn (e.g. health-monitoring shirts).
However their adoption remains limited, notably due to the limited considerations given to users' physical, psychological and social needs.
In this talk, I will present a few existing wearables and worldwide trends.
Then I will discuss needs in wearable computing, taking into account research on (1) human motivation, (2) childhood development and (3) changes in old age.

Speaker: Researcher at the 国立情報学研究所 (National Institute of Informatics), Dr Duval obtained his Ph.D. in information science from 総合研究大学院大学 (The Graduate University for Advanced Studies, Japan) in 2006.
His Ph.D. work was entitled: Satisfying Fundamental Needs in Everyday Life With Wearable Computers - The Case of Belonging Needs.
Dr Duval currently investigates specific needs in wearable computing from birth to old age, and leads a project dedicated to the support of families with ubiquitous technologies.

Website: http://horizons.free.fr/home/eng/projects.htm#psysoc-wearcomp

Date: June 14th (Thursday), 10:00-12:00am
Place: National Institute of Informatics, 12th floor, Lecture Room 1 (Room 1212)
Speaker:
Professor Zhong-Zhi Bai (Institute of Computational Mathematics and Scientific / Engineering Computing, Academy of Mathematics and Systems Science, Chinese Academy of Sciences)
Title:
On Hermitian and skew-Hermitian splitting iteration methods.

Abstract:
The Hermitian and skew-Hermitian splitting (HSS) iteration scheme is an efficient and practical method for solving large sparse non-Hermitian system of linear equations. In this talk, after reviewing the HSS iteration method and its basic convergence theory for non-Hermitian positive definite matrices, we give a sufficient　and necessary condition for guaranteeing its convergence for nonsingular and non-Hermitian positive　semidefinite matrices. We then discuss the semi-convergence property of the HSS iteration method and　derive a sufficient and necessary convergence condition for singular and non-Hermitian positive semidefinite　matrices. According to the optimal iteration parameter involved, we first compute it exactly for real　two-by-two matrices, and then compute it exactly for special block two-by-two matrices. These formulas　are used to give estimation for the optimal iteration parameter of HSS iteration method for general　non-Hermitian matrices. Finally, some numerical results are used to examine the effectiveness of the HSS　iteration method with the exact or the estimated optimal iteration parameter.

You are welcome to attend the following seminar.

Dr Kawazoe (NII) and Dr Doan (NII) will present their research on the annotation of multilingual biomedical texts, then Dr Apel (NII) will present his research on dictionary reversal for German-Japanese at the National Institute of Informatics.

Date: Thursday 14th of June 2007 (15:00-17:00)
Location: National Institute of Informatics　19F, Room 1904
Language: English
Registration fees:: None
Seminar page: http://horizons.free.fr/his/eng/seminars.htm

Semantic Annotation in BioCaster: its Design and Challenges
(15:00-15:30)
by Dr Ai Kawazoe

Abstract:
The BioCaster project aims to construct a text mining-based system that provides advanced search and analysis of disease outbreak reports on the Web for public health experts, clinicians and researchers interested in infectious diseases.
Its key component is the use of automatic learning methods to identify important entities and events using features derived from examples by human annotators. The nature of the task requires an expansion of "markable" categories of concepts, from those referred by proper nouns (names of person, organization, location) to those referred by common nouns and noun phrases, and also from context-independent concepts to concept-dependent ones (e.g. roles). We will present the design of annotation schema used for this project and discuss the difficult cases.

Speaker:
Dr Ai Kawazoe is a project researcher at the National Institute of Informatics since 2006. She studied theoretical linguistics and received her doctor's degree in literature from Kyushu university in 2005. Her current research focuses on the design of text annotation schema by applying formal ontological methodology and linguistic judgements.

Website: http://biocaster.nii.ac.jp/

The Roles of Named Entities and Their Roles in Classifying Annotated Biomedical Texts (15:30-16:00)
by Dr DOAN Son

Abstract:
This talk investigates the roles of named entities (NEs) in annotated biomedical text classification. In the annotation schema of BioCaster, a text mining system for public health protection, important concepts that reflect information about infectious diseases were conceptually analyzed with a formal ontological methodology. Concepts were classified as Types, while others were identified as being Roles.
Types are specified as NE classes and Roles are integrated into NEs as attributes. We focus on the Roles of NEs by extracting and using them as different features in the classifiers. We discuss in detail advantages of Roles in annotated biomedical texts. The effect of each Role on accuracy of text classification is also discussed.

Speaker:
Dr Doan is a project researcher at the National Institute of Informatics since 2006. He received PhD degree in computer science from Japan Advanced Institute of Science and Technology in September 2005, then moved to the graduated school of information science at Tohoku university as a post-doctor researcher. His interests are text mining, text categorization, information extraction, and machine learning.

Website: http://biocaster.nii.ac.jp/

Dictionary Reversal – Building a German-Japanese Dictionary from Japanese-German Data, Using Vocabulary Analysis, User Cooperation and Approaches from Natural Language Processing (16:00-17:00)
by Dr APEL Ulrich

Abstract:
With over 100,000 headwords and about 250,000 records the electronic WaDoku-Dictionary is the most comprehensive Japanese-German dictionary of its kind. Alone the possibility to search the German entries too doesn't make it a German-Japanese dictionary. In many cases a German query will result in too many found records, and it is difficult for the users to decide which Japanese term is the best in a certain case. Users might get for example obsolete Japanese entries, that are important for the sake of completeness of the Japanese side, but obviously shouldn’t be used when writing a text in modern Japanese. Users with Japanese native language will miss further information about the German entries like pronunciation, conjugation, declination, valency etc.

The dictionary reversal project will have several steps to develop more useful German-Japanese dictionary data. We will for example detect and mark-up entries that cannot or should not be reversed (e.g. definitions or archaic entries). We will mark entry domains, that are easily reversible (e.g. computer science, electronics, plant or animal names with scientific names), and we will mark domains that are problematic to reverse (e.g. Buddhist terms, traditional art, traditional medicine). We will further add existing data on German grammar, pronunciation etc., and revise data along a frequency list for German words.

The short-term aim is to build a usable comprehensive free German-Japanese dictionary in only a few months. This data will be improved through user cooperation as it is already implemented for the Japanese-German data.

Parts of our approach should be usable for the reversal of other dictionaries too, and by using it on free Japanese-English and English-Japanese dictionaries, we should be able to carry out even an evaluation of our approach.

Speaker:
Dr Apel studied Japanology, Sociology and Ethnology at Munich, Germany. He wrote his Ph.D. thesis on Japanese futures research and futures studies while staying at Osaka university. He works on the Japanese-German dictionary WaDoku-Jiten since 1998. From 2004 to 2005 he carried out research at the National Institute of Informatics with a JSPS scholarship. He continued research at the National Institute of Informatics as guest researcher, and since 2007 has the post of a project researcher there.

Website: http://wadoku.de/

Date: June 7(Thu) and 8(Fri), 2007
Venue: National Center for Sciences

You are welcome to attend the following talk.

Date: May 17th(Thursday), 10:00-12:00am
Place: National Institute of Informatics, 12F, Lecture Room 1 (1212)
Title: A Novel Multigrid Based Preconditioner For Heterogeneous Helmholtz Problems
Speaker: Professor Cornelis W. Oosterlee
(Delft University of Technology, Delft, the Netherlands, and, CWI, Center for Mathematics and Computer Science, Amsterdam.)

This is joint work with Y.A. Erlangga, C. Vuik, A. Kononov and Chr. Dwi Riyanti.

Abstract:
In this presentation an iterative solution method, in the form of a preconditioner for a Krylov subspace method, is presented for high wavenumber Helmholtz problems in heterogeneous media.
The preconditioner is based on the Helmholtz operator, where an imaginary term is added. This preconditioner can be handled by one iteration of the multigrid method. This may be somewhat surprising as multigrid, without enhancements, has convergence troubles for the original Helmholtz operator at high
wavenumbers. The choice of multigrid components for the corresponding preconditioning matrix with a complex diagonal is validated with Fourier analysis.
Multigrid analysis results are verified by numerical experiments. High wavenumber Helmholtz problems in 2D heterogeneous media are solved indicating the performance of the preconditioner. The method is parallelized and generalized to three dimensions. We will include a 3D example.

The research is financially supported by Dutch Ministry of Economic Affairs project BTS01044.

You are welcome to attend the following seminar.

Mr Picard (FT R&D) will present his research on the automatic construction of knowledge resources at France Telecom R&D and at the National Institute of Informatics, under the supervision of professor Aizawa.

Schedule: Thursday 17th of May 2007 (14:30-15:30)
Location: National Institute of Informatics (Tokyo), 20F, Room 2006
Language: English
Registration fees: None
Seminar page: http://horizons.free.fr/his/eng/seminars.htm

*Automatic Extraction from Web Corpora of Characteristic Properties of a Concept by Mr PICARD Etienne

Abstract:
The goal of our project is to automate the construction of knowledge resources. Our framework starts with a concept (e.g. museums, singers), and extracts from web corpora properties linking this concept to other ones. The initial concept is described by its name (e.g. museum) and by the name of instances (e.g. The Louvre). Based on this information, we use search engines to build texts corpora from the web, and apply text mining techniques to extract properties from those corpora. Our aim is to build resources usable by information retrieval dialoguing agents.

Speaker:
Mr Picard is a doctor student at Jospeh Fourier University of Grenoble (France). He conducts his research at France Telecom R&D's laboratory in Lannion (France), and joined the National Institute of Informatics for a 3-months internship under the supervision of AIZAWA Akiko. His current research topics deal with text mining, web content mining, and knowledge extraction.

You are welcome to attend the following seminar.

Associate Professor CHBEIR (University of Bourgogne, UMR-CNRS LE2I　Laboratory, France) will present his research on a XML similarity.

Schedule: Thursday 10th of May 2007 (15:00-16:00)
Location: National Institute of Informatics (Tokyo), 19F, Room 1904
Language: English
Registration fees: None

XML Similarity (15:00-16:00)
by Dr CHBEIR Richard

Abstract:
Similarity serves as an organization principle by which individuals classify objects, form concepts and make generalizations. It plays a central role in various research areas, particularly in the XML field where similarity evaluation of XML data has been receiving a lot of attention. In essence, W3C’s XML (eXtensible Mark-up Language) has recently gained unparalleled importance as a fundamental standard for efficient data management and exchange. Information destined to be broadcasted over the web is henceforth represented using XML, in order to guaranty its interoperability. The use of XML covers data representation and storage (e.g. complex multimedia objects), database information interchange, data filtering, as well as web services interaction. Owing to the ever-increasing abundant use of XML especially on the web, XML-based similarity/comparison becomes a central issue, specifically in the information retrieval (IR) and database (DB) communities, its applications ranging over:

• Version control, change management and data warehousing (finding, scoring and browsing changes between different versions of a document, support of temporal queries and index maintenance).


• Semi-structured data integration (measuring the similarity between XML documents in order to undertake the integration of corresponding data sources).


• Classification/clustering of XML documents gathered from the web against a set of DTDs declared in an XML database (just as schemas are necessary in traditional DBMS for the provision of efficient storage,
  retrieval, protection and indexing facilities, the same is true for DTDs and XML repositories).


• XML query systems (finding and ranking results according to their similarity in order to retrieve the best results possible).

In this talk, we give an overview of existing research related to XML similarity, in both its AI dynamic programming (ED-based approaches) and Information Retrieval fields. We show how:

• Most approaches in the ED literature focus exclusively on the structure of documents, ignoring the semantics involved.


• Most approaches ignore several XML similarity cases where the corresponding edit distance outcome is inaccurate.

We present our proposal and prototype aiming at both combining edit distance structural similarity computations with IR semantic similarity assessment, in an XML (structured data) context, and providing an improved fine-grained method for comparing heterogeneous XML documents.

Speaker:
Dr Chbeir received his Ph.D. in Computer Science from the University of INSA, France in 2000. He is currently Associate Professor of Computer Science in the University of Bourgogne, France. His research interests include multimedia information retrieval, distributed multimedia database management, spatio-temporal relations, access control models, bioinformatics, and the development and the integration of information systems. He is chair of ACM SIGAPP French Chapter, and member of several conference and journal program committees (EuroPar, IEEE ISSPIT, ACM ASIIS, ICIT, ACM SWS, etc.). He published in several international journals (e.g. IEEE Transactions on SMC, Journal on Data Semantics, Journal of Methods of Information in medicine) and conferences (e.g. IEEE SITIS, ACM SAC, Visual, IEEE, FLAIRS, IRMA).

You are welcome to attend the following seminar.

Dr HOULE (NII) will present his research on a *combinatorial approach to search and clustering*, partly carried out as joint work with Dr GRIRA (NII).

Schedule: Thursday 26th of April 2007 (15:00-17:30)
Location: National Institute of Informatics (Tokyo), 20F, Room 2005
Language: English
Registration fees: None
Seminar page: http://horizons.free.fr/his/eng/seminars.htm

A Combinatorial Approach to Search and Clustering (15:00-17:30)
by Dr HOULE Michael

Abstract:
This seminar introduces a new model for clustering based on combinatorial rather than distance information, that requires no direct knowledge of the nature or representation of the data. In lieu of such knowledge, the Relevant-Set Correlation (RSC) clustering model relies solely on the existence of an oracle that accepts a query in the form of a data item, and returns a ranked set of items relevant to the query. In principle, the role of the oracle could be played by any similarity search structure, or even a commercial search engine whose ranking function and relevancy scores are kept secret. The quality of cluster candidates, the degree of association between pairs of cluster candidates, and the degree of association between clusters and data items are all assessed according to the statistical significance of a form of correlation among pairs of relevant sets and/or candidate cluster sets. A clustering heuristic, GreedyRSC, has also been developed based on RSC. We show that GreedyRSC has many desirable features - in particular, it is able to determine an appropriate number of clusters naturally and automatically, without requiring the user to supply data-dependent parameter values.

Topics to be covered include:

• The Relevant-Set Correlation (RSC) model for clustering


• The GreedyRSC clustering heuristic


• Experimental results on large, high-dimensional data sets:




• Text


• Images


• Protein sequence




• Other applications of RSC:




• Query result clustering (joint work with Dr GRIRA Nizar, NII)


• Integration of clusterings


• Outlier detection (joint work with Mr GEBSKI Matthew, NICTA)


• Feature evaluation and selection (joint work with Dr GRIRA Nizar, NII)




• Efficient generation of relevant sets using the SASH approximate similarity search structure (if time and interest permits)

Speaker:
Dr Houle joined the National Institute of Informatics as a Visiting Professor in 2004. In addition to his current work on search and clustering for data mining applications, his research interests include computational and combinatorial geometry, data structures, and graph algorithms.

You are welcome to attend the following seminar.

Mr BECKER Christian (Universität Bielefeld, Germany) will present his doctoral research, and more precisely the *emotion simulation for a virtual human*. This work was partly carried out during a stay at NII under the supervision of Dr PRENDINGER Helmut (NII), sponsored by the JSPS.

Schedule: Wednesday 4th of April 2007 (15:00-16:00)
Location: National Institute of Informatics (Tokyo), 19F, Room 1904
Language: English
Registration fees: None
Seminar page: http://horizons.free.fr/his/eng/seminars.htm

*Emotion Simulation for a Virtual Human: Implementing the "As-If"
Body-Loop in a Cognitive Architecture (15:00-16:00)*
by Mr BECKER Christian

Abstract:
In recent years, the integration of emotion-driven behaviors became prominent in the field of virtual humans. The virtual human under development at the University of Bielefeld, called Max, is a testbed for studying human-like behavior in natural face-to-face interactions. To endow its cognitive architecture with simulated emotions, we follow Damasio’s distinction of "primary" and "secondary" emotions. Primary ones are elicited as an immediate reaction to a stimulus that might originate from internal bodily processes; secondary ones (e.g. relief, hope) arise from cognitively higher processes of conscious appraisal. I will first introduce our agent's cognitive architecture and its employment in two different interaction scenarios. Then I will present
the psychological background for emotion simulation, distinguishing structural and dimensional emotion theories. Finally I will outline the implementation of an "as-if" body-loop, based on a distinction between physis and cognition in our architecture to simulate and express primary and secondary emotions.

Speaker:
Mr Becker received his M.S. degree from the University of Bielefeld, and is currently a Ph.D. student and research assistant in its Artificial Intelligence Group. Pre-Doctoral Fellow of the Japan Society for the Promotion of Science (JSPS), he stayed at the National Institute of Informatics during 3 month in 2005. His current research focuses on the simulation of complex emotions for conversational artifacts.

Website: http://www.techfak.uni-bielefeld.de/~cbecker/index_engl.html

You are welcome to attend the following series of lectures.

Lecturer: Professor Michael Eiermann
Institut fuer Numerische Mathematik und Optimierung, Technische Universitaet Bergakademie Freiberg, Freiberg, Germany,
e-mail: eiermann [at] math.tu-freiberg.de

Title: Krylov Subspace Methods: Theory and Applications

Place: National Institute of Informatics 12F Lecture Room 1 (1212)

Dates:
2007
1) February 26th (Mon) 10:00-12:00 ( Introduction.)
2) February 27th (Tue) 10:00-12:00 ( Projection methods on expanding subspaces.)
3) February 28th (Wed) 10:00-12:00 ( Coordinate representation and algorithms.)
4) March 1st (Thu) 10:00-12:00 ( Krylov subspaces.)
5) March 5th (Mon) 15:00-17:00 ( Arnoldi-based Krylov subspace methods.)
6) March 6th (Tue) 15:00-17:00 ( The conjugate gradient (CG) method.)
7) March 8th (Thu) 10:00-12:00 ( Lanczos-based Krylov subspace methods.)
8) March 9th (Fri) 10:00-12:00 ( Convergence.)
9) March 12th (Mon) 10:00-12:00 ( Practical issues.)
10) March 14th (Wed) 10:00-12:00 ( Krylov methods for singular problems.)
11) March 16th (Fri) 10:00-12:00 ( Krylov methods for matrix functions.)

Abstract ＞＞PDF

Most tasks in scientific computing ultimately boil down to the
solution of systems of linear equations. Discretizations of
differential or integral equations usually result in systems of
algebraic equations. When these equations are nonlinear they have
to be linearized, e.g., by Newton's method, and finally we face
the question: What is the solution of A x = b ?

Many linear systems that arise in practical problems (especially
those which result from finite element or finite difference
discretizations of partial differential equations) can be so huge
that limited storage space as well as limited computing time
generally prohibit the application of direct solvers such as
Gaussian elimination. Fortunately, large `real-life' matrices are
often sparse, i.e., they have only a few nonzero entries. For
practical purposes this means that matrix-vector products with A
can be computed cheaply. Iterative methods generate a sequence of
approximate solutions, where the main computational effort for
constructing the m-th approximant from the previous one consists
in one or a few matrix-vector multiplications with A, and this
is why large and sparse systems are usually solved iteratively.

Krylov subspace methods, the topic of these lectures, form the
most important class of iterative solution method. In the past
three decades research on Krylov subspace techniques has brought
forth a variety of algorithms and methods so large that even
specialists in matrix computations have difficulties keeping up.

It is our objective to develop the theory and algorithms on which
all Krylov subspace methods are based in a unified way, to
emphasize their connections to other fields of applied mathematics
(such as polynomial approximation), but to treat also problems one
encounters in practise, e.g., their behavior in finite precision
arithmetic and how their convergence can be accelerated using
preconditioners. In addition, we shortly describe how Krylov
subspace methods help to solve other large linear algebra problems
such as finding a few eigenpairs of a matrix, reducing the
dimension of a linear model or evaluating a matrix function.

1. Introduction.
Model problems. Sparse matrices and iterative methods. Krylov subspace
methods and preconditioning. The method of successive approximation and
other classical iterative methods.

2. Projection methods on expanding subspaces.
Minimal residual (MR) and orthogonal residual (OR) subspace correction.
Projections and angles. Projections onto nested subspaces. MR and OR
approximations on nested subspaces. Relations between nested MR
and OR approximations.

3. Coordinate representation and algorithms.
Working with coordinates. The orthogonalization process. Angles
and the QR-factorization. The Paige-Saunders basis. Using
arbitrary bases. Quasi-minimal and quasi-orthogonal
approximations. Multiple subspace correction.

4. Krylov subspaces.
Why Krylov subspaces? Shift operator, orthogonal and kernel
polynomials, Gaussian quadrature. Parameterization of the Arnoldi
process. Short recurrences.

5. Arnoldi-based Krylov subspace methods.
A minimal residual method (GMRES). The full orthogonalization
method (FOM). Restarted and truncated variants. GMRES for
Hermitian systems: MINRES.

6. The conjugate gradient (CG) method.
Some remarks on the history of CG. Different views on CG. Gaussian
quadrature and the CG method. CG convergence. CG and normal
equations.

7. Lanczos-based Krylov subspace methods.
The nonsymmetric Lanczos process and look-ahead strategies.
Lanczos-based equation solvers. The quasi-minimal residual methods
(QMR). The biconjugate gradient method (BiCG). BiCGStab and other
product methods.

8. Convergence.
Linear versus superlinear convergence. Bounds based on angles
between subspaces. Convergence analysis for the normal case.
Convergence results based on potential theory. Convergence
analysis for the nonnormal case. Bounds based on the field of
values. Bounds based on pseudospectra.

9. Practical issues.
Krylov subspace methods in finite precision. Error estimates and
stopping criteria. Choice of the initial approximation.
Preconditioning.

10. Krylov methods for singular problems.
The minimal polynomial and the Drazin inverse. Termination of
Krylov subspace methods. Least-squares solutions in Krylov spaces.
Krylov subspace methods for the Drazin inverse solution.

11. Krylov methods for matrix functions.
The definition of matrix functions. Krylov subspace approximations
for matrix functions. Algorithms. Convergence.

Date: February 9th (Friday) 11:00-12:00

Place: National Institute of Informatics 20F Seminar Room (2006)

Speaker: Professor Siegfried M. Rump, Institute for Reliable Computing, Hamburg University of Technology (http://www.ti3.tu-harburg.de)

Title: Inversion of extremely ill-conditioned matrices in double precision

Abstract:
It is well known that the maximum meaningful condition number of a matrix to
be inverted in a floating point arithmetic with relative rounding error unit
EPS is about 1/EPS. For IEEE 754 double precision this limits the condition
number to about 10^16.

In a recent paper we showed that using only ordinary double precision the
sum and dot product of vectors can be computed accurately to the last bit,
independent of the condition number of the sum or dot product. In this talk
we show that using this we can invert matrices of arbitrary condition
number, for example 10^50, in ordinary double precision.

Our algorithm is old, but was never published because the lack of analysis.
Now we can present an analysis of the algorithm giving new and unexpected
insight into floating point arithmetic. The bottom line is that rounding
into floating point is the key to success.