Department of Mathematics

Search

Optimization seminar

×

Modal title

Modal content

Autumn Semester 2011

Date / Time Speaker Title Location
26 September 2011
16:30-18:00 		Dr. Wolfram Wiesemann
Imperial College London, London, UK
Event Details

Optimization Seminar

Title Robust Markov Decision Processes
Speaker, Affiliation Dr. Wolfram Wiesemann, Imperial College London, London, UK
Date, Time 26 September 2011, 16:30-18:00
Location HG G 19.1
Abstract Markov decision processes (MDPs) are powerful tools for decision making in uncertain dynamic environments. However, the solutions of MDPs are of limited practical use due to their sensitivity to distributional model parameters, which are typically unknown and have to be estimated by the decision maker. To counter the detrimental effects of estimation errors, we consider robust MDPs that offer probabilistic guarantees in view of the unknown parameters. To this end, we assume that an observation history of the MDP is available. Based on this history, we derive a confidence region that contains the unknown parameters with a pre-specified probability 1-beta. Afterwards, we determine a policy that attains the highest worst-case performance over this confidence region. By construction, this policy achieves or exceeds its worst-case performance with a confidence of at least 1-beta. Our method involves the solution of tractable conic programs of moderate size.
Robust Markov Decision Processesread_more
HG G 19.1
3 October 2011
16:30-18:00 		Prof. Dr. Elad Hazan
Technion - Israel Institute of Technology, Haifa, Israel
Event Details

Optimization Seminar

Title Sublinear optimization for machine learning
Speaker, Affiliation Prof. Dr. Elad Hazan, Technion - Israel Institute of Technology, Haifa, Israel
Date, Time 3 October 2011, 16:30-18:00
Location HG G 19.1
Abstract Linear classification is a fundamental problem of machine learning, in which positive and negative examples of a concept are represented in Euclidean space by their feature vectors, and we seek to find a hyperplane separating the two classes of vectors. We give the first sublinear-time (information-optimal) algorithms for linear classification, support vector machine training, and other related problems in machine learning, including the kernelized versions of these problems. These new algorithms are based on a primal-dual approach, and use a combination of novel sampling techniques and the randomized implementation of online learning algorithms. We give lower bounds which show our running times to be nearly best possible in the unit-cost RAM model. Joint work with Ken Clarkson and David Woodruff, appeared in FOCS 2010. The talk is self-contained, no prior knowledge in machine learning is assumed.
Sublinear optimization for machine learningread_more
HG G 19.1
10 October 2011
16:30-18:00 		Alexander Meeraus
GAMS Development Corporation, Washington, USA
Event Details

Optimization Seminar

Title CANCELLED: Mathematical Programming Modeling Systems: Past - Present - Future
Speaker, Affiliation Alexander Meeraus, GAMS Development Corporation, Washington, USA
Date, Time 10 October 2011, 16:30-18:00
Location HG G 19.1
Abstract Some 30 years ago, the Research Center at the World Bank embarked on a series of projects to improve the productivity and reliability of applying operations research tools to problems in economic and project planning. An experimental large scale implementation of the emerging concepts of relational algebra and automatic differentiation finally lead to the development of the General Algebraic Modeling System (GAMS) in collaboration with industrial partners. Looking back at issues and problems that motivated the development of GAMS do not look so different from problems we face today.
CANCELLED: Mathematical Programming Modeling Systems: Past - Present - Futureread_more (CANCELLED)
HG G 19.1
17 October 2011
16:30-18:00 		Prof. Dr. Giovanni Puccetti
University of Firenze, Italy
Event Details

Optimization Seminar

Title Computation of sharp bounds on the distribution of a function of dependent risks
Speaker, Affiliation Prof. Dr. Giovanni Puccetti, University of Firenze, Italy
Date, Time 17 October 2011, 16:30-18:00
Location HG G 19.1
Abstract We introduce a new algorithm to compute numerically sharp lower and upper bounds on the distribution of a function of d dependent random variables having fixed marginal distributions. In quantitative risk management, the bounds can be used to compute best-possible bounds on Value-at-Risk for portfolios of dependent risks. Compared to the existing literature, the bounds are widely applicable, more accurate and more easily obtained. (joint work with Ludger Lüschendorf)
Computation of sharp bounds on the distribution of a function of dependent risksread_more
HG G 19.1
31 October 2011
16:30-18:00 		Prof. Dr. Michael C. Ferris
University of Wisconsin, Madison, USA
Event Details

Optimization Seminar

Title Multiple Optimization Problems with Equilibrium Constraints
Speaker, Affiliation Prof. Dr. Michael C. Ferris, University of Wisconsin, Madison, USA
Date, Time 31 October 2011, 16:30-18:00
Location HG G 19.1
Abstract We present a mechanism for describing and solving collections of optimization problems that are linked by equilibrium conditions. Included in this class are classical models such as the PIES model and agent based formulations arising from Nash Games. We demonstrate this mechanism in the context of energy planning problems, specifically for capacity expansion, hydro operation, and demand response. We believe that optimization is part of a larger process, wherein optimization processes can be used to develop inputs for other parts of the solution (data processing), its strong theory (of duality for example) can enable more effective solution schemes to exploit model structure, and the interplay between collections of models treating stochastic effects, competition between independent agents, and a mixture of continuous and discrete approaches can provide richness for describing, solving and adapting the underlying system being modeled. Mechanisms to allow decision making in such circumstances can be informed by game theory and techniques from distributed computing. The extended mathematical programming (EMP) framework exists to provide access to modeling formats that fall outside the classical framework. We give examples of their use and their possible extensions to hierarchical modeling.
Multiple Optimization Problems with Equilibrium Constraintsread_more
HG G 19.1
7 November 2011
16:30-18:00 		Prof. Dr. Martin Henk
Otto-von-Guericke-University of Magdeburg, Germany
Event Details

Optimization Seminar

Title On Feasibility of Integer Knapsacks
Speaker, Affiliation Prof. Dr. Martin Henk, Otto-von-Guericke-University of Magdeburg, Germany
Date, Time 7 November 2011, 16:30-18:00
Location HG G 19.1
Abstract Given a matrix A \in Zmxn satisfying certain regularity assumptions, we consider the set F(A) of all vectors b \in Zm such that the associated knapsack polytope P(A, b) = {x \in Rn : Ax = b, x>=0} contains an integer point. When m=1 the set F(A) is known to contain all consecutive integers greater than the so called Frobenius number associated with A. Here we introduce a kind of multi-dimensional Frobenius number F(A) which reflects in an analogous way feasibility properties of the problem and the structure of F(A) in the general case. We give an optimal upper bound for F(A) and also estimate the asymptotic growth of the diagonal Frobenius number on average. We further discuss complexity problems related to F(A). (This is joint work with Iskander Aliev)
On Feasibility of Integer Knapsacksread_more
HG G 19.1
14 November 2011
16:30-18:00 		Prof. Dr. Rüdiger Kiesel
Universität Duisburg-Essen, Germany
Event Details

Optimization Seminar

Title Electricity Forward Markets
Speaker, Affiliation Prof. Dr. Rüdiger Kiesel, Universität Duisburg-Essen, Germany
Date, Time 14 November 2011, 16:30-18:00
Location HG G 19.1
Abstract In this talk we provide frameworks to explain the market risk premium in electricity forward markets, defined as the difference between forward prices and spot forecasts. We show how it depends on the risk preferences of market players and what impact information differences may have. We will perform an empirical investigation of the so-called information premium, which is defined as the influence of future information not incorporated in spot prices but taken into consideration when pricing forwards. Additionally, we will consider the problem of optimal capacity allocation to forward and spot markets. (joint work with Fred Espen Benth, Richard Biegler-Koenig and Katrin Jensen)
Electricity Forward Marketsread_more
HG G 19.1
21 November 2011
16:30-18:00 		Prof. Dr. Rainer Tichatschke
University of Trier, Germany
Event Details

Optimization Seminar

Title On proximal-like methods for equilibrium problems
Speaker, Affiliation Prof. Dr. Rainer Tichatschke, University of Trier, Germany
Date, Time 21 November 2011, 16:30-18:00
Location HG G 19.1
Abstract The relation between fixed point problems and equilibrium problems allows us to reformulate (by means of the Nikaido-Isoda-function) a Nash equilibrium problem into a corresponding variational inequality with a pseudo-monotone operator. Therefore, the usage of proximal like regularization methods becomes appropriate for their numerical solution. Using zone-coercive Bregman-like functions, the generated regularized auxiliary problems may be treated as unconstrained ones. As a sort of inexactness, we allow both: The usage of ε-enlargements of the operator as well as the inexact solution of these auxiliary problems. This increases the numerical attraction of the method suggested. Shortly we discuss the convergence of the methods (the whole theory works without para-monotonicity assumptions) and show some numerical experiments with benchmark examples. Key words: Saddle point problems, Nash equilibria, pseudo-monotone variational inequalities, proximal-like regularization, Bregman distances.
On proximal-like methods for equilibrium problemsread_more
HG G 19.1
28 November 2011
16:30-18:00 		Prof. Dr. George Constantinides
Imperial College London, UK
Event Details

Optimization Seminar

Title FPGA Acceleration of Embedded Numerical Algorithms
Speaker, Affiliation Prof. Dr. George Constantinides, Imperial College London, UK
Date, Time 28 November 2011, 16:30-18:00
Location HG G 19.1
Abstract This talk will discuss the potential of Field-Programmable Gate Array (FPGA) architectures to accelerate numerically intensive applications, with a particular emphasis on structured quadratic programming problems arising in model predictive control. After a brief tutorial overview of FPGAs, we discuss their advantages and challenges for this application area, including a delve into some details of hardware implementations of computer arithmetic. We then provide an overview of our current implementation of FPGA-based MPC, presented at FPGA 2011 in Monterey, before discussing some of the open questions and challenges that remain. The work presented has been jointly developed with Eric Kerrigan and Juan Jerez.
FPGA Acceleration of Embedded Numerical Algorithmsread_more
HG G 19.1
5 December 2011
16:30-18:00 		Prof. Dr. Asen Dontchev
University of Michigan, Ann Arbour and American Mathematical Society, USA
Event Details

Optimization Seminar

Title Implicit Functions, Metric Regularity, and Convergence of Algorithms
Speaker, Affiliation Prof. Dr. Asen Dontchev, University of Michigan, Ann Arbour and American Mathematical Society, USA
Date, Time 5 December 2011, 16:30-18:00
Location HG G 19.1
Abstract The classical implicit function theorem revolves around solving an equation f(p; x) = 0 for x in terms of a parameter p; and tells us when the solution mapping associated with this equation is a differentiable function with respect to the parameter. In this talk we move into a much wider territory in replacing equation-solving problems by more complicated problems for "generalized equations" that arise in constrained optimization, models of equilibrium, control theory, and many other areas. It turns out that if we put aside differentiability and focus on Lipschitz continuity only, or even more general metric regularity properties of mappings, we can cover a wider range of models and get estimates of the solution changes resulting from approximations of the model. After a review of metric regularity properties of set-valued mappings, we present an application to inexact Newton's method for solving variational inequalities. We also show an extension of the Dennis-Moré theorem for generalized equations.
Implicit Functions, Metric Regularity, and Convergence of Algorithmsread_more
HG G 19.1
19 December 2011
16:30-18:00 		Prof. Dr. Antonio J. Conejo
Universidad de Castilla - La Mancha, Ciudad Real, Spain
Event Details

Optimization Seminar

Title Pricing Non-convexities in an Electricity Pool
Speaker, Affiliation Prof. Dr. Antonio J. Conejo, Universidad de Castilla - La Mancha, Ciudad Real, Spain
Date, Time 19 December 2011, 16:30-18:00
Location HG G 19.1
Abstract Electricity pools are generally cleared through auctions that are conveniently formulated as mixed-integer linear programming problems. Since a mixed-integer linear programming problem is non-continuous and non-convex, marginal prices cannot be easily derived. However, to trade electricity, prices are needed. Thus, a relevant question arises: how to generate appropriate prices? This presentation addresses this important issue and proposes a primal-dual approach to derive efficient pool-clearing prices that support market outcomes in the sense that agents are willing to remain in the market. Such prices do not significantly deviate from the marginal prices obtained if integrality conditions are relaxed in the original mixed-integer linear programming problem. Two case studies illustrate the functioning of the proposed pricing scheme.
Pricing Non-convexities in an Electricity Poolread_more
HG G 19.1

Note: if you want you can subscribe to the iCal/ics Calender.

Organizers: Komei Fukuda, Bernd Gärtner, Diethard Klatte, Hans-Jakob Lüthi, John Lygeros, Manfred Morari, Karl Schmedders, Robert Weismantel

Archive: SS 22 SS 20 AS 19 SS 19 AS 18 SS 18 AS 17 SS 17 AS 16 SS 16 AS 15 SS 15 AS 14 SS 14 AS 13 SS 13 AS 12 SS 12 AS 11 SS 11 AS 10 SS 10 AS 09

JavaScript has been disabled in your browser