Informatica

We apply some concepts of Information-Based Complexity (IBC) to global and discrete optimization. We assume that only partial information on the objective is available. We gather this partial information by observations. We use the traditional IBC definitions and notions while defining formal aspects of the problem. We use the Bayesian framework to consider less formal aspects, such as expert knowledge and heuristics.

In this paper optimization aspects relatively to circuit component placement problem for gate array VLSI are discussed. Practical and theoretical aspects of the methods of component placement are concerned as well. Effective heuristic algorithms for the initial placement and iterative placement improvement are described. An original strategy of global placement optimization is investigated. Some experimental results based on an automatic placement subsystem for gate arrays – AUTOPLACE developed at Department of Practical Informatics of Kaunas University of Technology are presented.

We consider finite population slotted ALOHA where each of n terminals has its own transmission probability pi. Given the overall traffic load λ, the probabilities pi are determined in such a way as to maximize throughput. This is achieved by solving a constrained optimization problem. The results of Abramson (1970) are obtained as a special case. Our recent results are improved (Mathar and Žilinskas, 1993).

In this paper we consider the numerical solution of the large system of nonlinear differential equations. We assume that the system simulates the semiconductor circuit. We apply the well known event driven techniques to get some approximation of solution fast. We extend those techniques by considering pairs of nodes, instead of single nodes, as usual. The “pairwise” solution is more efficient in tightly coupled circuits. The improvement of efficiency of solution is important optimizing the parameters of circuits. In many cases we need the global optimization of equation parameters. Then, the modeling speed could be the main factor of successful application.

We consider here the optimization problems of simple competitive model. There are two servers providing the some service. Each server fix the price and the rate of service. The rate of service defines the customer losses waiting in line for the service. The customer go to the server with lesser total service cost. The total cost includes the service price plus waiting losses. A customer goes away, if the total cost exceeds some critical level. The flow of customers and the service time both are stochastic. There is no known analytical solution for this model. We get the results by Monte Carlo simulation. We get the analytical solution of the simplyfied model.

We consider here the average deviation as the most important objective when designing numerical techniques and algorithms. We call that a Bayesian approach.

The problem we are dealing with is following. There exist certain number of nodes η, transmitting messages at random time moments. If time interval between messages transmitted by different nodes are less than some given value, a collision occurs. We can fix the collision, but we cannot determine the nodes engaged in the collision. The hierarchical decomposition of the nodes is used to resolve the collision. At every hierarchical level, a subset of nodes “suspected” as participating in the collision is divided in a certain number of groups. There is a time period given to every group, at which messages can be transmitted. This proceeds while no more collisions occurs. This paper covers the problems of selecting a number of groups, to minimize the longest collision resolution time, as well as average collision resolution time.

The paper deals with the minimization algorithms which enable us to economize the computing time during the coordinated calculation of the values of an objective function on the nodes of a rectangular lattice by storing and using quantities that are common for several nodes. The algorithm of a uniform search with clustering, the variable metric algorithm and the polytope algorithm are modified.