Madison Chaos and Complex Systems Seminar

Fall 1996 Seminars

Dates, speakers, titles and abstracts will be listed as they become available. Meetings will be noon Tuesdays in 4274 Chamberlin Hall unless otherwise noted.

Short List

3 September. Steering Committee Meeting

Speaker nomination and recruitment. Funding. Possible themes for series of talks.

10 September. Bill Lytton, UW Neurology. ``The Thalamic Oscillator: Complex Cell or Simple Network?''

Abstract: Neurons are complex dynamical systems with multiple internal interactions that are presumably relevant to their information processing role. We have used computer modeling to investigate the neurons of the thalamus, a central nuclear complex that serves both as a way-station for sensory input and as a central oscillator during sleep. Three main cell types make up the thalamic network: thalamocortical cells (TC), reticularis cells (RE) and thalamic interneuron (INT). Depending on membrane voltage, each neuron type can by itself produce oscillatory behavior. Connected together, they influence each other's phase and frequency to produce the oscillatory patterns of the thalamus as a whole. 

17 September. Jean-Paul Chavas, UW Agricultural Economics. ``Dynamics and Economic Rationality.''

Abstract: Economic rationality and efficient resource allocation are typically defined in the context of optimizing behavior. This raises the question: Under what conditions does optimizing behavior generate simple dynamics (e.g., convergence to a steady-state equilibrium)? And under what conditions does it generate complicated dynamics (e.g., limit cycles or chaos)? Implications are discussed concerning the formulation and solution of dynamic optimization problems, the explanation of business cycles, and the characterization of economic rationality. 

24 September. J. A. Scott Kelso, Center for Complex Systems, Florida Atlantic University. ``How Things Work Together: The Problem of Coordination.''

Abstract: Coordination is a fundamental feature of living things, implying some kind of functional order in which the ``whole is greater than and different from the parts''. Historically, ideas from Gestalt psychology, Bernstein's synergy concept, Weiss's notion of self-differentiation, von Holst's principle of central coordination are all attempts to come to grips with the coordination problem. Only recently have these intuitions been provided a firm theoretical and empirical foundation. Building upon the concepts and methods of pattern formation in nonequilibrium systems, coordination dynamics aims to find the basic rules for how coordination states form and change. The first part of this talk reviews some research in which it has been possible to identify relevant coordination variables and their dynamics for several different kinds of system (e.g. parts of the organism, the organism responding to signals from its environment, organisms interacting with each other). Importantly, predictions of the theory have been subject to experimental tests, a process that has led to new theoretical developments. The second part of the talk will describe how the strategy of coordination dynamics, in which instability plays a key conceptual and methodological role, has been extended to the study of how parts of the brain are coordinated with each other and with behavior itself. Work is underway that attempts to derive earlier phenomenological models of brain and behavioral function from a deeper theory grounded in current knowledge about cortical anatomy and neurophysiology. If there is time, I will talk about what all this might mean for the coordination of mind, brain and behavior.

1 October. W. A. Brock, UW Economics. ``Evolutionary Theories of the Stock Market.''

Abstract: This talk gives a tour of recent efforts to modify the standard rational expectations literature to include expensive acquisition of rational expectations. For financial models fitness of an expectation or belief system is past trading profits net cost of acquisition. The distribution of fitnesses changes over time. There are periods where markets run ``tight'' and periods where they run ``loose.'' This is so because if everyone buys expensive but more accurate expectations then the profit is competed away to the extent that the acquisition cost is not covered and vice versa. This tension can create complicated dynamics. A discussion will be given of problems that deterministic dynamical systems models face in generating patterns that look like real stock market data. 

8 October. Chris Demarco, UW Electrical and Computer Engineering. ``Predicting Dynamic Behavior in Large Scale Electric Power Networks: the Anatomy of Blackout.''

Abstract: The North American electric power network is composed of hundreds of thousands of electromechanical devices, coupled by a transmission grid of continental scale. The composite behavior of this huge interconnected system can be highly complex. This behavior is further complicated by the unavoidable occurrance of indivdual equipment faliures, and the presence of an overlaid protection system that can instantaneously disconnect additional pieces of equipment when local overloads are detected. This talk will give an overview of typical disturbance scenarios in such systems, taking the recent western US blackouts of July and August '96 as motivating examples. We will continue by providing an overview of the basic structure of the underlying electromechanical dynamics, and describe how these interact with protective devices in a manner that may be modeled as a ``discrete event'' system. To efficiently operate a modern power grid, it is clearly desirable to be able to predict the dynamic response of the system to anticipated disturbances. This talk will briefly survey the current state of the art in simulation tools for this purpose, and will highlight current research efforts at UW-Madison that seek to augment these with analytic tools for assessing qualitative behavior. 

15 October. Rick Jenison, UW Psychology. ``Dynamics Bayesian Estimation of Time-to-Arrival from Acoustic Information.''

Abstract: Much of the interest regarding an active observer has centered on precise timing of interactions with objects in the environment in relation to observers, for example, a baseball player catching a fly ball. An observer seeks new information by exploring the environment. This means moving around, tracking objects in space, and avoiding collisions. These examples certainly seem natural for the visual sense, but what about audition? Although the physical nature of optics and acoustics are different, both kinds of information can be shown to sufficiently specify the geometry or kinematics of moving objects relative to an observer.

Theoretical work on optic information for time-to-arrival has focused on a variable commonly referred to as ``tau'' that specifies the time-to-direct-contact from the optical dilation of an approaching object at any instant in time. An analysis similar to that for visual tau has been applied to acoustic structure to derive an analogous acoustic tau based on detected changes in sound intensity. Change in sound intensity is but one source of information for specifying motion parameters of a sound-emitting object in space. Changes in binaural and spectral structure in the forms of dynamic interaural-time-delay, interaural-intensity-differences, and Doppler shifts are also available. In this talk, a dynamical system for integrating acoustic information that specifies time-to-arrival (and parameters of object motion in general) is presented. This model is based on Bayesian data-fusion techniques that have been developed in the field of passive sonar. Although these measurement equations are nonlinear, they can be linearized sufficiently to utilize standard Kalman filtering techniques as recursive estimators of motion-state variables. Theoretical lower bounds can be established using the Cramer-Rao lower bound on the recursive estimators, which gives the minimum variance that an unbiased estimator can achieve from a series of noisy measurements.

22 October. Bob Savit, Department of Physics, University of Michigan. ``Time Dependence in Complex Systems.''

Abstract: Time series generated by complex, nonlinear systems are notoriously difficult to analyze. One important problem that must be addressed early on, is whether the underlying dynamics driving the process is time independent. This question is related to, but not the same as, the question of stationarity. In this talk I will describe a new method for analyzing time series from complex systems that is designed to uncover time dependence in the underlying dynamics. In addition to addressing the question of whether the underlying dynamics is time dependent or not, the approach provides insight into the nature of the time dependence, if it exists. I will discuss applications of the method to the analysis of intracranial EEG recordings from patients with temporal lobe epilepsy, to a problem in mechanical vibration analysis, as well as to some examples of simulated data.

29 October. David Albers, UW Physics. ``Dynamical Behavior of Artificial Neural Networks with Random Weights.''

Abstract: In this talk I will describe a Monte Carlo study of large untrained, feedforward, neural networks with randomly chosen weights and feedback. The analysis consists of looking at the percent of the systems that exhibit chaos, the distribution of largest Lyapunov exponents, and the distribution of correlation dimensions. As the systems become more complex (increasing inputs and neurons), the probability of chaos approaches unity. The correlation dimension is typically much smaller than the system dimension.

5 November. Richard K. Belew, CSE Dept. --- UC San Diego and CS Dept. --- UW (Visiting). ``Competitive Co-evolution.''

Abstract: Many of the most important design problems are difficult not only because there are a vast number of solutions to consider, but also because the number of TESTS for potential solutions is vast as well. We analyze the ``competition dynamic'' between solutions and tests as it arises in evolutionary computations: Two distinct populations of ``hosts'' and ``parasites'' are maintained, with increasing reproductive success of individuals in one population being at the expense of those in the other. The resulting search is related to the theory of PAC learning, and several heuristics are demonstrated to significantly improve the evolutionary algorithm's search behavior. Game-playing applications are shown to be a very natural domain for these methods, and some prelimary results applied to the board game of Go are presented.

This work is based on an upcoming dissertation by Chris Rosin, CSE/UCSD.

11 November. Related Lecture. John Lisman, Volen Center for Complex Systems, Brandeis University. ``The Role of Brain Oscillations in Long term and Short Term Memory.''

Unusual time and place: Monday, 3:30 pm, room 341 Bardeen.

Sponsored by the Center for Neuroscience (Ken Mack, 263-9800, or Peter Lipton, 262-1709, for more information or contact with Dr. Lisman).

12 November. Robert Meyer, UW Computer Sciences. ``A Genetic Algorithm for Optimal Domain Decomposition.''

Abstract: The problem of ``optimally'' partitioning a given domain into a specified number of pieces arises in a variety of computer science applications, from database to image analysis to partial differential equations. The general goal in all of these applications is to obtain pieces that are of nearly equal size (for load balancing in parallel computing environments) and, in addition, have small perimeter (in the two-dimensional case) or small surface area (in the three-dimensional case), thereby reducing communication costs. This type of problem fits well into the genetic algorithm paradigm since it can be formulated in terms of trying to determine and combine good ``building blocks'' (well-shaped pieces of the domain) to obtain a good overall solution. We present a genetic algorithm that outperforms existing approaches to domain decomposition because it makes effective use of this high-level building block information.

15 November. Related Lecture. Stuart Zola: ``Brain Circuitry of Memory and Memory Loss: Findings from Humans and Nonhuman Primates.''

Unusual time and place: Friday, 12 noon in room 113 Psychology.

A University Lecture sponsored by the Psychology Department and co-sponsored by the Primate Center and the Department of Psychiatry. ``Dr. Zola is an internationally renowned researcher on neural mechanisms of memory formation and memory loss, whose work spans both animal models and humans, using a variety of manipulative and imaging methodologies.'' Contact for information: Craig Berridge, Psychology Dept.

15 November. Related Lecture. Josh Chover, UW Math. ``Sequential Recall.''

Abstract: I'll discuss a simple two-layer neural network model which has sparse forward, lateral, and feedback connections. Simulations demonstrate an ability for quick recall of a sequence of overlapping inputs, given an initial prompt--inputs learned in an unsupervised way. The dynamics are not based on iteration to a limit. The model suggests roles for biological counterparts, such as modifiable and non-modifiable channels.

19 November. Paul Terry, UW Physics. ``Impeding Turbulent Transport: From Fusion Reactors to the Ozone Hole.''

Abstract: There is no universal theory of turbulence, given the lack of universality in the dynamical behavior of different systems. However a recently discovered phenomenon runs counter to this trend and appears to operate in essentially the same fashion in systems ranging from fusion plasmas to the earth's middle atmosphere. This phenomenon, suppression of turbulence and turbulent mixing by sheared flow, has been credited with removing turbulent transport in fusion devices. It may also apply to transport barriers in the stratosphere, confinement of salinity and heat in long-lived vortical ocean currents, and intermittency in 2D Navier-Stokes turbulence. This phenomenon will be described heuristically and its application to some of the above problems will be discussed.

26 November. George E. Hrabovsky, UW Physics. ``Transient Chaos in Accretion Disks.''

Abstract: This talk will discuss transient chaos in accretion disks (disks of matter which spiral in to a source of intense gravitation such as black holes and neutron stars). A review of topics involving circle maps, transient chaos, diffusion, and accretion will be a part of the talk.

26 November. Related Lecture. Subbarao Palacharla, UW Computer Sciences. ``Quantifying the Complexity of Superscalar Processors.''

Unusual Time and Place: 4pm, room 1325 Computer Sciences building.

Abstract: This talk consists of two parts. The first part describes our attempt at quantifying the complexity of superscalar processors. First, a generic superscalar pipeline is defined. Then, the specific areas of register renaming, window wakeup and selection logic, and operand bypassing are analyzed. Performance (delay) results and trends are expressed in terms of issue width and window size. Our analysis indicates that window logic and operand bypass logic are likely to be the most critical in the future.

The second part of the talk describes a microarchitecture that simplifies window logic. This dependence-based microarchitecture puts dependent instructions into multiple queues, and issues them from the queues in strict FIFO order. Simulation shows little slowdown as compared with a completely flexible issue window, when performance is measured in clock cycles. Furthermore, because only instructions at FIFO heads need to be awakened and selected, issue logic is simplified and the clock cycle is faster -- consequently overall performance is improved. Also, the proposed microarchitecture, by grouping dependent instructions together, helps minimize the performance degradation due to slow bypasses in future wide-issue machines.

3 December. Virginia Young, UW School of Business. ``Creating an Expert System from Linguistic Rules--A Fuzzy Logic Approach.''

Abstract: Fuzzy logic was developed in the 1970's to provide a way to account for the vagueness inherent in linguistic rules. I will give a brief background into fuzzy sets themselves and will describe how one can use fuzzy logic to create an expert system by starting with linguistic rules. I will illustrate my method with a simple hypothetical example from insurance pricing. 

10 December. Christopher Kribs, UW Math. ``A Two-Sex STD Model with Recruitment and Activity Levels.''

Abstract: In this talk I will present a brief overview of dynamical systems that have been used to model the spread of a disease within a population, and then discuss a particular model I have developed which incorporates division of the population both by gender and by contact rate (here interpreted as sexual activity levels), separations which normally have not been modeled simultaneously because of the inherent complexity of the model(s). Analysis of the model yields results which differ from those of simpler models: in particular, the coexistence of two stable equilibria, so that local stability does not imply global stability. The physical assumptions for the model are based on studies of sexually transmitted diseases with eventual recovery (e.g., gonorrhea). My emphasis will be on results and their implications rather than their derivations.
Up to the Chaos and Complex Systems Seminar page.
Last change worth mentioning Mon Nov 25 10:46:05 1996