BEGIN:VCALENDAR
METHOD:PUBLISH
PRODID:-//Apple Computer\, Inc//iCal 1.0//EN
X-WR-CALNAME;VALUE=TEXT:USC
VERSION:2.0
BEGIN:VEVENT
DESCRIPTION:Speaker: Valentino Crespi,
Talk Title: Trackability and Machine Learning of Processes
Abstract: The effective monitoring of complex environments is related to the ability of machine learning and tracking its constituent processes.\n
Examples of environments in this domain include networked computer systems, autonomic computing systems and distributed and dynamic information systems. In our approach an environment consists, in its most abstract form, of multiple processes or behaviors that we typically model as Finite State Machines such as Probabilistic and nonprobabilistic Finite State Automata (DFAs/PFAs), Probabilistic Deterministic Finite State Automata (PDFAs), Probabilistic Suffix Automata (PSAs), Hidden Markov Models (HMMs), etc.\n
\n
In this talk we first introduce an original and rigorous concept of "trackability" of processes in a distributed sensing system. The purpose of this notion is to determine the "complexity" of estimating state trajectories of a target process based on a discrete-time sequence of noisy "observations". We then present our new algorithms to machine learn Hidden Markov Models (HMMs) from typical realizations of the associated stochastic process. The methods are based on the non-negative matrix factorization (NMF) of higher order Markovian statistics and are structurally different from the classical Baum-Welch and associated approaches.\n
Host: Bhaskar Krishnamachari
SEQUENCE:5
DTSTART:20110202T103000
LOCATION:EEB 248
DTSTAMP:20110202T103000
SUMMARY:EE-Systems Seminar
UID:EC9439B1-FF65-11D6-9973-003065F99D04
DTEND:20110202T113000
END:VEVENT
END:VCALENDAR