トップ   差分 バックアップ リロード   一覧 単語検索 最終更新   ヘルプ   最終更新のRSS

**第9回 [#c3e57bee]
日時 	2009年3月18日(水) 15:30-16:30
**第8回 [#v0b4806e]

会場 	西8号館W棟8階W1008号室
日時: ''2010年3月31日(水) 11:00 - 12:00''

講演者 	高橋渉教授(東京工業大学)
会場: ''西8号館W棟8階W809号室''

題目 	Proximal Point Algorithms and Four Nonlinear Resolvents in Optimization
講演者: ''Marielba Rojas 氏 (Delft University of Technology, The Netherlands)''

**第8回 [#cd5c219e]
日時 	2009年3月5日(木) 15:00-16:30
題目: '' Accelerating the LSTRS Algorithm''

会場 	西8号館W棟8階W1008号室

講演者 	吉田真紀氏(大阪大学)
The LSTRS algorithm presented in M. Rojas, S.A. Santos, and D.C. Sorensen.
A new matrix-free method for the large-scale trust-region
subproblem, SIAM J. Optim. 11(3): 611-646, 2000 is
an iterative procedure for solving large-scale quadratic problems
with one norm constraint, or trust-region subproblems. The method is based
on a reformulation of the trust-region subproblem as a parametric eigenvalue
problem. The main computation in LSTRS is the solution of a large, symmetric
eigenvalue problem at each step.

題目 	情報セキュリティ研究生活と大阪大学における男女共同参画の取り組み
The associated software (M. Rojas, S.A. Santos, and D.C. Sorensen.
Algorithm 873: LSTRS: MATLAB software for large-scale trust-region subproblems
and regularization, ACM Trans. Math. Software :11, 2008) offers several
possibilities for the solution of the eigenproblems.

We present a brief description of the LSTRS algorithm with focus
on the eigenvalue problems and their features. We describe recent work
on improving the performance of LSTRS by means of efficient non-linear
techniques for eigenvalue computations.

This is joint work with Joerg Lampe, Danny Sorensen, and Heinrich Voss.

**第7回 [#sd17f2ff]
日時 	2009年2月19日(木) 14:00-15:00
**第7回 [#j0c03368]

会場 	西8号館W棟8階W809号室
日時: ''2010年3月30日(火) 16:00 - 17:00''

講演者 	Sunyoung Kim 教授(Ewha W. University)
会場: ''西8号館W棟10階W1008号室''

題目 	A Sparse Version of Full Semidefinite Programming Relaxation for Sensor Network Localization Problems
講演者: ''Andrej Bogdanov 氏 (The Chinese University of Hong Kong)''

題目: ''Adaptivity helps in testing bipartiteness''

Semidefinite programming relaxation has become an important tool for sensor network localization after Biswas and Ye's work. We discuss a MATLAB software package called "a sparse version of full semidefinite programming relaxation (SFSDP)." SFSDP is proposed to improve the efficiency of the full semidefinite programming relaxation by Biswas and Ye by exploiting the sparsity of the problem. Thus, the size of the resulting SDP relaxation is reduced and much larger-sized problems can be handled. Numerical results show better computational performance of SFSDP than any other available methods.

At the end of the talk, the status of women in science and engineering in Korea will be discussed briefly, as a part of promoting women in science program.
We say a graph is epsilon-far from bipartite if it cannot be made
bipartite even after removing an arbitrary set of epsilon n^2 of its
edges. A test for bipartiteness (with distance parameter epsilon) is a
randomized algorithm that, given access to the adjacency matrix of a
large graph G, queries a small number of entries in the matrix,
accepts if G is bipartite and rejects  with high probability if G is
epsilon-far from bipartite.

**第6回 [#u2ccc0ba]
We give a new algorithm that tests bipartiteness in time
O((1/epsilon)^{2-c}), where c > 0 is some universal constant. Our
algorithm is inherently adaptive, as it is known that non-adaptive
algorithms for testing bipartiteness require Omega((1/epsilon)^2)
queries. (Based on joint work with Li Fan)

日時 	2008年12月12日(金) 15:00-16:00
**第6回 [#c362b7af]

会場 	西8号館W棟10階W1008号室
日時: ''2010年3月29日(月) 16:30 - 17:30''

講演者 	Professor Anthony To-Ming Lau (University of Alberta)
会場: ''西8号館W棟8階W809号室''

題目 	Separation and extension properties for positive definite functions for groups
講演者: ''高橋 幸雄(数理・計算科学専攻)''

**第5回 [#d14f344e]

A well known theorem of Hahn Banach asserts that if E is a closed subspace of a Banach space and F is a closed subspace of E, then each continuous linear functional on F can be extended to a continuous linear functional on E. As a consequence, any element in E but not in F can be separated by continuous linear functional on E. In my talk, I shall discuss similar properties for closed subgroups of a locally compact group and their relations with the invariant complementation problem for the group von Neumann algebra of a locally compact group.
日時: ''2010年2月25日(木) 第1部 15:00 - 15:50, 第2部 16:00 - 17:00''

**第5回 [#dad1fd37]
会場: ''東工大蔵前会館 手島精一記念会議室 (L, S)''

日時 	2008年10月22日(水) 15:30-17:30
講演者: ''吉田 真紀 氏 (大阪大学)''

会場 	西8号館W棟10階W1008号室
講演者 	古宇田悠哉(Yuya KODA)
第1部 ''多様性か否か - 安全性の形式的検証'', 
第2部 ''多様性か否か - 男女共同参画''

題目 	Colored Turaev-Viro invariants of twist knots
暗号理論のなかでも最近注目を集めている安全性の形式的検証に ついてお話しいただきます。また、大阪大学における女性研究者 支援の取り組みについてもご紹介いただきます。

文部科学省科学技術振興調整費「理工系女性研究者 プロモーションプログラム」の助成を受けています。

至るところが3次元ユークリッド空間と同相な近傍を持つ空間を3次元多様体という.向き付けられた3次元多様体$M$と,その中に埋め込まれた有向絡み目$L$,すなわち向きが指定されたいくつかの円周のイソトピー類,の対$(M,L)$に対し,色付きTuraev-Viro不変量とよばれる不変量が Barrett,Garcia-Islas,Martinsの3氏により導入された.この不変量は,イニシャルデータとよばれるある代数的な情報と,これに付随して与えられる絡み目の"色"が与えられるごとに定義されるものであり, 3次元多様体$M$とその中の絡み目$L$の対$(M,L)$に対して"真の意味"で定義される不変量である.本講演では,ツイスト結び目とよばれる3次元球面内の結び目のクラスに対するこの不変量の計算方法と計算結果を紹介する.時間が許せば,レンズ空間内の同種のクラスに対する計算結果も紹介する.
**第4回 [#v136b20a]

2 講演者 	安永憲司(Kenji YASUNAGA)
日時: ''2010年1月15日(金) 16:30 - 17:30''

題目 	誤り訂正符号における誤りの単調性を利用した訂正能力分析」
"Analysis of error correcting codes using the monotone error structure"
会場: ''西8号館W棟10階W1008号室''

講演者: ''Wei-Ming Ni 氏(客員教授,University of Minnesota)''

最小距離がdである誤り訂正符号を使えば、d/2個未満の誤りは必ず訂正することができる。 d/2個以上の誤りが発生した場合でも誤り個数がd/2付近であれば、多くの誤りが訂正可能であることが知られている。しかし、この問題に対する理論的分析はそれほど進んでいるとはいえない。講演者はこれまでに、d/2個以上の誤りが発生した場合の訂正可能誤り数について、誤りの単調性を利用した分析を行ってきた。本講演ではその成果について説明する。

**第4回 [#rc28845b]
日時 	2008年10月1日(水) 13:30-14:30
要旨: In this talk I will discuss and compare the following two aspects of the dynamics of 2x2 reaction-diffusion equations and their shadow systems: global existence/finite-time blow-up, and, stability properties of steady states. A well known activator-inhibitor system in morphogenesis developed by Alan Turing , and A. Gierer and H. Meinhardt, will be used as a primary example to illustrate the results.

会場 	西8号館W棟10階W1008号室

講演者 	Professor Suhyoung Choi (KAIST)
**第3回 [#kbd5e04e]

題目 	A survey of projective geometric structures on 2-, 3-manifolds
日時:   ''2009年11月25日(水) 16:30 - 17:30''

会場:   ''西8号館W棟10階W1008号室''

We go over definitions of classical geometries including projective geometry. We discuss manifolds with geometric structures on them and their significance. Finally, we discuss manifolds with projective structures and many recent advances and some questions that we are working on using algebraic and numerical computations.
講演者:  ''Hans D. Mittelmann 氏 (Arizona State University)''

**第3回 [#kd0fc26a]
日時 	2008年9月1日(月) 13:30-14:30
題目:   ''The Power of SDP Relaxations --- Computing Strong Bounds for QAPs and Graph Problems''

会場 	西8号館W棟8階W809号室
As is well-known semidefinite relaxations of discrete optimization
problems yield excellent bounds on their solutions. We present two
examples from our recent research. The first addresses the quadratic
assignment problem and a formulation is developed which yields the
strongest lower bounds known for a number of large cases (up to
dimension 256). It utilizes standard SDP software. With iterative
SDP solvers still larger dimensions as they occur, for example,
in communications, are within reach. The second area is the
computation of bounds in graph problems, respectively in geometry.
A strategy based on the Lovasz theta function is generalized to
compute lower bounds for the chromatic number of graphs and upper
bounds on the spherical kissing number utilizing SDP relaxations.
Multiple precision SDP solvers are needed and improvements on known
results for kissing numbers in dimensions up to 23 are obtained.

講演者 	Dr. Aapo Hyvärinen (Professor of Computational Data Analysis, University of Helsinki, Finland)

題目 	From independent component analysis to score matching
**第2回 [#q938586b]

日時:   ''2009年6月3日(水) 13:30 - 14:30''

I will start the talk by reviewing the basic concepts of independent component analysis (ICA), and its application to learning Bayesian networks. Generalizations of ICA serve as one motivation for the main topic of the talk: estimation of probabilistic models which are not normalized, i.e. they do not integrate into unity.
会場:   ''西8号館W棟10階W1008号室''

Normalization is, in principle, always possible by numerical integration, but it is often computationally very difficult. To solve this problem, I introduced a new estimation principle: score matching (JMLR, 2005). The point is to look at some derivatives of the log-densities which do not depend on the normalization of the model at all. The objective function is given in closed form and can be readily minimized by conventional optimization techniques, thus avoiding any numerical integration. The ensuing estimation method can also be shown to have some statistical optimality properties in Bayesian denoising, and leads to a new kind of information geometry.
講演者: ''Joseph Maher 氏 (Oklahoma State University)''

題目:   ''Surfaces, groups and random walks.''

**第2回 [#xd15ad2a]

日時 	2008年7月9日(水) 13:30-14:30
Every group has an intrinsic geometry associated with it, which can help us understand properties of the group, such as:
- what does the group looks like on the large scale?
- what properties does a generic group element have?

会場 	西8号館W棟10階W1008号室
We will consider such questions for various families of groups, such as free  groups, matrix groups, and groups arising from the study of surfaces, and we will mention some applications to 3-manifolds. 

講演者 	Paul Dow 氏
**第1回 [#k8e2f936]

題目 	Recent Progress in Finding Exact Treewidth
日時: ''2009年5月7日(木) 13:20-''

会場: ''西8号館W棟8階W809号室''

The treewidth of a graph is a scalar value that measures how easily a graph can be decomposed into a tree. Many NP-hard problems can be solved in polynomial time when restricted to graphs with bounded treewidth. Additionally, treewidth is central to several active areas of artificial intelligence research, including probabilistic reasoning and constraint satisfaction. In these research areas, the ability to find the exact treewidth of a graph can have a significant impact on the size and difficulty of problems that can be solved. Unfortunately, finding the treewidth of a general graph is NP-hard, and it is not known whether a polynomial-time, constant-factor approximation is possible. In this talk, I will discuss new techniques that have allowed us to find the treewidth of ever larger and more difficult graphs. The techniques I will discuss come from the area of heuristic search algorithms, including best-first search, breadth-first heuristic search, iterative deepening, and the use of external memory via delayed duplicate detection.
講演者:	''Karl Sigman 氏 (Columbia University)''

**第1回 [#g1257dba]
題目: ''Brief introduction to Monte Carlo Simulation and its use for estimating the price of simple stock options (under the Binomial Lattice Model)''

日時 	2008年5月12日(月) 13:30-14:30

会場 	西8号館W棟10階W1008号室

講演者 	Manindra Agrawal 氏 (Indian Institute of Technology Kanpur)

題目 	Basic Tools of Computational Algebra


Algorithms for a number of problems of both theoretical and practical importance have been designed with the help of algebra. For example, coding messages, cryptography, image compression, integer factorization, primality testing, polynomial factorization etc. Nearly all of these algorithms use only basic tools of algebra. In this talk, I will give an overview of some of these and discuss their usage.