|
|
Risk-Sensitive Optimality in Markov Games
Sladký, Karel ; Martínez Cortés, V. M.
The article is devoted to risk-sensitive optimality in Markov games. Attention is focused on Markov games evolving on communicating Markov chains with two-players with opposite aims. Considering risk-sensitive optimality criteria means that total reward generated by the game is evaluated by exponential utility function with a given risk-sensitive coefficient. In particular, the first player (resp. the secondplayer) tries to maximize (resp. minimize) the long-run risk sensitive average reward. Observe that if the second player is dummy, the problem is reduced to finding optimal policy of the Markov decision chain with the risk-sensitive optimality. Recall that for the risk sensitivity coefficient equal to zero we arrive at traditional optimality criteria. In this article, connections between risk-sensitive and risk-neutral Markov decisionchains and Markov games models are studied using discrepancy functions. Explicit formulae for bounds on the risk-sensitive average long-run reward are reported. Policy iteration algorithm for finding suboptimal policies of both players is suggested. The obtained results are illustrated on numerical example.
|
|
|
Scenario Generation via L-1 Norm
Kaňková, Vlasta
Optimization problems depending on a probability measure correspond to many economic and financial situations. It can be very complicated to solve these problems, especially when the underlying probability measure belongs to continuous type. Consequently, the underlying continuous probability measure is often replaced by discrete one with finite number of atoms (scenario). The aim of the contribution is to deal with the above mentioned approximation in a special form of stochastic optimization problems with an operator of the mathematical expectation in the objective function. The stability results determined by the help of the Wasserstein metric (based on the L_1 norm) are employed to generate approximate distributions
|
|
|
Second Order Optimality in Transient and Discounted Markov Decision Chains
Sladký, Karel
The article is devoted to second order optimality in Markov decision processes. Attention is primarily focused on the reward variance for discounted models and undiscounted transient models (i.e. where the spectral radius of the transition probability matrix is less than unity). Considering the second order optimality criteria means that in the class of policies maximizing (or minimizing) total expected discounted reward (or undiscounted reward for the transient model) we choose the policy minimizing the total variance. Explicit formulae for calculating the variances for transient and discounted models are reported along with sketches of algoritmic procedures for finding second order optimal policies.
|
|
|
Statistical analysis of competing risks in an unemployment study
Volf, Petr
This study is concerned with the analysis of dependence of random variables - latent times to events, in a competing risks case. We discuss first the problem of identifiability of marginal and joint distributions of competing random variables. Then, the copula models are utilized in order to express the dependence. Finally, the Gauss copula is used to solution of a real example with unemployment data.
|
|
|
A note on the use of copulas in chance-constrained programming
Houda, Michal
In this paper we are concentrated on a problem of linear chanceconstrained programming where the constraint matrix is considered random with a known distribution of the matrix rows. The rows are not considered to be independent; instead, we make use of the copula notion to describe the dependence of the matrix rows. In particular, the distribution of the rows is driven by so-called Archimedean class of copulas. We provide a review of very basic properties of Archimedean copulas and describe how they can be used to transform the stochastic programming problem into a deterministic problem of second-order cone programming. Also the question of convexity of the problem is explored and importance of the selected class of copulas is commented. At the end of the paper, we provide a simple example to illustrate the concept used.
|
|
|
Multiobjective Stochastic Optimization Problems with Probability Constraints
Kaňková, Vlasta
Rather general multiobjective optimization problems depending on a probability measure correspond often to situations in which an economic or financial process is simultaneously influenced by a random factor and a “decision” parameter; moreover simultaneously it is reasonable to evaluate the process by a few objective functions and it seems reasonable to determine the decision with to the mathematical expectation of objectives. A complete knowledge of the probability measure is a necessary assumption to analyze the problem. However, in applications mostly the problem has to be solved on the data base. A relationship between “characteristics” obtained on the base of complete knowledge of the probability measure and them obtained on the above mentioned data base has been already investigated in the case when constraints are not depending on the probability measure. The aim of the talk will be to relax this condition.
|
|
| |
|
|
On Bayes approach to optimization
Volf, Petr
In many real optimization problems we have not full information on the objective function and can afford to evaluate it at just a few points. Then, certain assumptions on the objective function must be done. This could be taken as a prior information in a Bayes scheme. The Bayes approach to optimization then offers the way of effective search for the extremal point. We describe the technique how to mapproach the optimum using the Gauss process or a regression-like models.
|
|
|
On quantile optimization problem with censored data
Volf, Petr
In the framework of stochastic optimization the criterion based on selected quantiles is considered. Further, stochastic characteristics are estimated from censored data. Therefore, certain theoretical results concerning estimators of distribution function and quantiles under censoring are recalled and utilized to prove consistency of solution based on estimates. Behavior of solutions for finite data sizes is studied with the aid of randomly generated example.
|
|
|
Economic and Financial Problems via Multiobjective Stochastic Optimization
Kaňková, Vlasta
Multiobjective optimization problems depending on a probability measure correspond to many economic and financial activities. Evidently if the probability measure is completely known, then we can try to influence economic process employing methods of multiobjective deterministic optimization theory. Since this assumption is fulfilled very seldom we have mostly to analyze the mathematical model and consequently also economic process on the data base. The aim of the talk will be to investigate a relationship between ``characteristics" obtained on the base of complete knowledge of the probability measure and them obtained on the above mentioned data base. To this end, the results of the deterministic multiobjective optimization theory and the results obtained for stochastic one objective problems will be employed.
|
guest ::
