SEMINARS AND EVENTS IN FALL 2001.

1. Alan Mishchenko November 30.
   
   Expressing Flexibility of MV Networks using Partial Cares
   
   ABSTRACT.
   
   An MV function is a function, in which the inputs and the output take values from a finite set. An MV function can be completely specified (if each minterm of the input space has exactly one output value), with don't-cares (if the function can take any value in some minterms of the input space), or with partial cares (if the function takes an arbitrary subset of values in each minterm). The partial cares are the most general way of expressing incomplete specification of MV functions.
   
   An MV network is a directed acyclic graph, with nodes represented by MV functions. The sources of the graph are called primary inputs; the sinks are called primary outputs. The functionality of a MV network can be seen as a multi-output MV function giving the values of the primary outputs if the values at the primary inputs are known. The flexibility of a node in the network describes the limits, within which the node's functionality can be changed without influencing the functionality of the network.
   
   This talk discusses the use of partial cares to express several types of flexibility of MV nodes in the MV network. The complete flexibility describes the limits within which the node can be changed if the functionality of the other nodes is fixed. The compatible flexibility assigned to all the nodes describes the limits within which all nodes can be changed simultaneously. Efficient ways to represent and compute the partial cares are discussed. Numerical results show the amount of flexibility extracted using different algorithms and the degree of simplification of the nodes achieved using this flexibility.
   
   About the speaker:
   
   Alan Mishchenko graduated with honors from Moscow Institute for Physics and Technology, Moscow, Russia (1993), and received his Ph.D. in Computer Science from Glushkov Institute of Cybernetics, Kiev, Ukraine (1997). Since 1998, he has been a visiting scientist and an Intel sponsored researcher at PSU. Alan's research interests include formal methods of logic synthesis and verification, CAD tools, decision diagrams and their applications, and EXOR logic.
   
   
2. Martin Lukac - December 7. A New Approach to Problem Solving based on Artificial Life ideas.
   
   ABSTRACT.
   
   From Neural Nets point of view, there are two types of learning: supervised and unsupervised. This definition can be extended with the concept of self-learning or learning from data under personal constraints. The purpose of this presentation is to explore the adaptation of a cellular automaton to its environment. We based our research on dynamic programming, especially on TD approach coupled with a self-learning process. Using Artificial Life, theory we provide our algorithm with a reward function so as it is able to maximize its gains and change its behavior. The method seems to be as general as fuzzy logic or evolutionary approaches proposed before.
   
   In the presented approach we construct a matrix with random distribution of food and poison, and we measure the automaton's life time. Paradoxically, as evidenced by early experiments, the automata's survival is maximal in very bad conditions. Although we do not use any symbol manipulation techniques yet, we believe that this kind of learning can be generalized to any kind of problems due to its general aspects. In the future we plan to apply this method to any type of problems such a logic cicuit synthesis or mobile robot decisions making.
   
   About the speaker:
   
   Martin Lukac is a Ph.D. student in Department of Electrical and Computer engineering at Portland State University. His research interests are in artificial life, complex systems, learning and self-learning, and application of biological models to engineering. He has obtained his B.Sc degree in cellular biology and M.S. in Cognitive Sciences from Polytechnic School, France. He is associated with Intelligent Robotics Laboratory at PSU and a supervisor of his Ph.D. is Dr. Perkowski.