Talk Titles and Abstracts (tentative)
Brian Anderson, Australian National University
Title: Determining Causality
Abstract:
A control systems problem addressed several decades ago was to determine from measurements on different parts of a system whether there was feedback present in the system or not. Such problems as it turned out were of very great interest to economists, who studied this sort of question intensively. The name of Nobel-winning economist, Clive Granger, is part of the term Granger Causality, which is a cohesive body of ideas in stochastic processes, relevant to treating the question. More recently, such questions have arisen in theoretical and experimental studies in functional neuroimaging, which can attempt to find directional pathways in the brain.
This talk introduces a number of examples of causality and then reviews the definition of Granger causality and several characterizations of it. Granger causality is related to, but not identical with, physical causality. Then recent joint work with M Deistler and J.-M. Dufour is reviewed, examining the effect of measurement noise, measurement filtering and subsampling of measured signals on conclusions of a Granger causality nature.
Paul Fuhrmann, Ben-Gurion University of the Negev
Title: Thoughts on Optimal Control
Abstract:
Link to PDF
Tryphon T. Georgiou, University of California, Irvine
Title: On the Geometry of Optimal Mass Transport,
Where Probability, Control, and Physics Meet
Abstract:
With the backdrop of two closely related topics,
Monge-Kantorovich Optimal Mass Transport (OMT)
and Schroedinger's Bridges (entropy regularized OMT),
we will revisit links to Stochastic and Covariance Control,
and expand on the relevance of the Geometry of OMT in
modeling as well as in regulating stochastically
driven dynamical systems. We will underscore the
significance of the theory to problems in particle
systems, thermodynamics, and the dynamics of open
quantum systems.
The talk is based on joint works with
Yongxin Chen (GaTech), Giovanni Conforti
(Ecole Polytechnique), Wilfrid Gangbo (UCLA),
Michele Pavon (University of Padova),
and Allen Tannenbaum (Stony Brook).
Alberto Isidori, University of Rome
Title: Output Regulation of
Multivariable Invertible Nonlinear Systems
via Robust Dynamic Feedback Linearization
Abstract:
Since a long time, it has been known that multivariable nonlinear
systems, if invertible, can be input-output linearized via dynamic
extension and state feedback. The design methods in question though,
suffer severe limitations: they need access to the full state of the
controlled plant and they are intrinsically non-robust, being based on
exact cancellations. Recent developments have shown how, for special
classes of nonlinear systems, such limitations can be overcome. In
particular, the use of the so-called extended observers has proven
instrumental in context. Very recently, a design technique based on the
interlaced use of extended observers and dynamic extension has been
developed, by means of which robust asymptotic feedback linearization
can be obtained for a reasonably general class of invertible
multivariable systems. In the present paper, it will be shown how the
technique in question can be successfully used to solve also a problem
of robust output regulation.
Pramod Khargonekar, University of California, Irvine
Title: Signal Reconstruction, Neuroscience,
and Deep Learning
Abstract:
Anders Lindquist, Shanghai Jiao Tong University
Title: Dynamic Relations in Sampled Processes
Abstract:
In this talk we present some recent joint work with Tryphon Georgiou. Linear dynamical relations that may exist in continuous-time, or at some natural sampling rate, are not directly discernable at reduced observational sampling rates. Indeed, at reduced rates, matrix-valued spectral densities of vector-valued time series have maximal rank and thereby cannot be used to ascertain potential dynamic relations between their entries. This hitherto undeclared source of inaccuracies appears to plague off-the-shelf identification techniques seeking remedy in hypothetical observational noise. In this talk we explain the exact relation between stochastic models at different sampling rates and show how to construct stochastic models at the finest time scale that data allows. We then point out that the correct number of dynamical dependences can only be ascertained by considering stochastic models at this finest time scale, which in general is faster than the observational sampling rate. Thus, the principle contribution of this work is to introduce the idea of lifting identified models to a finer time-scale before assessing their complexity.
Stephen Morse, Yale University
Title: Estimating the State of
a Linear System Across a Network
Abstract:
The problem of estimating the state of a linear system
whose measured outputs are distributed across a network
has been under study in one form or another for a number
of years. Despite this, only recently have provably
correct distributed state observers emerged
which solve this problem under reasonably non-restrictive
assumptions. The aim of this talk is to describe
some of these observers and the conditions under
which they can provide asymptotically correct state
estimates.
Yoshito Ohta, Kyoto University
Title: Implication of the Lifting
Technique to System Identification
Abstract:
The celebrated lifting technique plays an instrumental role in sampled data control where a continuous-time system with a sampler and a hold is controlled by a discrete-time controller. It enables to design a feedback controller that considers intersample behavior unlike a controller based on a zero-order hold equivalent. This can be interpreted as a continuous-time to discrete-time system transformation using a shift-invariant subspace and its orthogonal complement.
This technique has the implication beyond sampled data control. This talk summarizes some topics on system identification using an extension of the lifting technique. The system transformation can be extended to include a noise process, which is applied to continuous-time system identification. A basis defined by the shift-invariant space such as Laguerre and Kautz is exploited in Bayesian system identification where the kernel is decided by orthonormal basis functions. These observations suggest that the idea of the lifting technique has far-reaching impact beyond just sampled-data control.
Hitay Ozbay, Bilkent University
Title: Interpolation-based Design of
Stabilizing Controllers for Retarded and
Neutral Delay Systems
Abstract:
A class of retarded and neutral time delay systems are considered: the
plant may have finitely many or infinitely many unstable poles.
Stabilizing controllers are obtained from a particular interpolation,
which leads to a finite dimensional sensitivity function. Extension to
more general systems with multiple state and input delays are shown. A
parameterization of all stabilizing integral action controllers are
obtained. Examples are given to illustrate this simple design
procedure. (joint work with Nazli Gundes)
Malcolm Smith, University of Cambridge
Title: Adjustable Lossless Mechanisms:
What Is It Possible to Build?
Abstract:
Is it possible to build a spring with a "workless knob" which freely adjusts its stiffness in real time? Such a contrivance would behave like a conventional linear spring when the knob is stationary. Energy imparted through compression or extension would be available for extraction again. Adjustment of the knob would not involve any energy transfer between the environment and the contrivance. Current methods to adjust the stiffness of springs do not answer this question, since they require active actuation, dissipation, or restrictive conditions on the switching of the spring constant. The talk is based on recent work with Tryphon Georgiou and Faryar Jabbari and will provide an answer to this and related questions.
Toshiharu Sugie, Osaka University
Title: Kernel Methods for Data-driven Control
Abstract:
Kernel methods, which play an important role in machine leaning, have
attracted a lot of attention in the area of system identification
recently. Once an appropriate kernel function is chosen, it is possible
to obtain a fairly accurate finite impulse response model of the target
system based on the short I/O data only. First, we show that the kernel
based identification is useful to update the feedforward compensation in
model matching problem. Its effectiveness is demonstrated through an
experiment of quadcopter position control. Second, using a Gaussian
kernel, we try to tune the PID controller based on the experimental
data. Its effectiveness is demonstrated through experiments of a
nonlinear mechanical system.
Mathukumalli Vidyasagar, Indian Institute of Technology Hyderabad
Title: Ramanujan Graphs and the Matrix Completion
Problem
Abstract: