Conferences, Lectures, & Seminars
Events for September
-
Learning Syntax and Semantics of Descriptions of Shapes from Web-users in the Smart-Bees Project
Tue, Sep 08, 2009 @ 03:30 PM - 05:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Sergio Guadarrama, Ph.D.,Fundamentals of Soft-Computing Unit
European Centre for Soft Computing
Mieres, SPAINAbstract: Smart-Bees is a project to study how machines can learn and communicate in human-like ways, from a Computing with Words (CW), Actions and Perceptions (CW-AP) perspective. Due to initial stage of development of CW and its complexity, setting up good experiments that can guide the research is crucial; so, following Wittgenstein's ideas about Meaning and Language ("The meaning of a word is its use in Language" and "Language games") we have developed a web-based application for automatically creating natural language-based descriptions of shapes. Smart-Bees is a multi-agent, multi-user system in which multiple smart-bees (agents or synthetic characters) are able to interact and communicate within themselves and with humans; and, where each smart-bee can learn concepts, words and also some sort of syntax and semantics, building a model that is grounded in perceptions and actions. The system learns from the descriptions provided by humans and then uses computational intelligence based methods and fuzzy algorithms to produce its own descriptions. To collect human descriptions of shapes and to test the results of Smart-Bees we have set up an interactive website, which will be described.Bio: Sergio Guadarrama received a PhD degree (Artificial Intelligence) in 2007, from the Technical University of Madrid. During 2003-2004, he spent one year at the University of California, Berkeley, where he collaborated with Prof. Lotfi Zadeh. He has published eight papers in international journals, seven book chapters and more than 20 papers in peer-reviewed conferences. He has participated in the organization of several international conferences, organized special sessions, and has also participated in several research projects funded by European Union, Spanish Government, Madrid Regional Government and Castilla la Mancha Regional Goverment. In recognition of his work, he has received: the "Best Doctoral Dissertation Award" in Computer Science by the Technical University of Madrid, and the "Juan de la Cierva" Grant, as one the top ten most promising young researchers in Computer Science by the Spanish Department of Science and Innovation. His main research focus is in Computing with Words, Actions and Perceptions (CW-AP), in which he is currently developing the project "Smart-Bees: Learning Language in Semi-Supervised Environments". Since November 2007, he is a "Juan de la Cierva" Researcher in the Fundamentals of Soft-Computing Unit at the European Center for Soft-Computing.Hosted by Professor Jerry Mendel
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248
Audiences: Everyone Is Invited
Contact: Talyia Veal
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
A Class of Restless Bandit Problems: Indexability and Optimality of Whittles Index
Wed, Sep 16, 2009 @ 02:00 PM - 03:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Dr. Qing Zhao
Assoc. Professor
U.C. Davis
Abstract:
The long‐standing multi‐armed bandit problem (MAB), propounded in early 1930's, was
solved by Gittins almost 40 years later in late 1960's when he established the simple
index structure of the optimal policy. In 1988, Whittle generalized MAB to the so‐called
restless multi‐armed bandit problem (RMAB) to take into account system dynamics that
cannot be directly controlled. Gittin's index policy is no longer optimal, and RMAB has
been shown to be PSPACE‐hard in general.
In this talk, we present a brief history of bandit problems and some recent results on a
special class of RMAB. We show that this class of RMAB is indexable, and Whittle's index
policy has a simple semi‐universal structure and achieves optimality.
This class of RMAB is particularly relevant to cognitive radio, user/server scheduling,
and optimal activation in multi‐agent systems.
Bio:
http://www.ece.ucdavis.edu/~qzhao/
Qing Zhao received the Ph.D. degree in Electrical Engineering in 2001 from Cornell
University, Ithaca, NY. In August 2004, she joined the Department of Electrical and
Computer Engineering at UC Davis where she is currently an Associate Professor.
Her research interests are in the general area of dynamic systems and communication
networks. She received the 2000 Young Author Best Paper Award from IEEE Signal Processing
Society and the 2008 Outstanding Junior Faculty Award from the UC Davis College of
Engineering.
Host: Bhaskar KrishnamachariLocation: Hughes Aircraft Electrical Engineering Center (EEB) - 248
Audiences: Everyone Is Invited
Contact: Shane Goodoff
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
Failures of Iterative Decoders
Thu, Sep 17, 2009 @ 02:00 PM - 03:00 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Shashi Kiran Chilappagri,
University of ArizonaAbstract: Low-density parity-check (LDPC) codes have been the focus of much research in the last decade due to their capacity approaching performance even when decoded with low-complexity algorithms. The message passing algorithms to decode LDPC codes are iterative in nature and work on the Tanner graph representation of the code. While these algorithms are optimal when the underlying graph is a tree, they exhibit surprisingly good performance on loopy graphs. The sub-optimality of these algorithms, however, manifests in the low noise regime leading to a abrupt performance degradation known as the error floor.In this talk, we provide an overview of the recent developments in understanding the error floor phenomenon. For the special case of the Gallager A algorithm, we analyze the trapping sets (the sub-graphs that lead to decoding failures) and present a combinatorial characterization of the underlying topological structure of these sub-graphs. We present tight upper and lower bounds on the guaranteed error correction capability of column-weight-three codes under the Gallager A algorithm. We also show how the knowledge of trapping sets can be used for performance estimation as well as code design.Biography: Shashi Kiran Chilappagri obtained his B.Tech and M.Tech degrees from the Indian Institute of Technology, Madras in 2004 and Ph.D. from the University of Arizona, Tucson in 2008, all in electrical engineering. He is currently a Research Engineer at the University of Arizona. His research interests include coding theory, graph theory as well as application of coding theory to inference problems in biology.Host: Alex Dimakis, dimakis@usc.edu, EEB 532, x09264
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 539
Audiences: Everyone Is Invited
Contact: Gerrielyn Ramos
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor. -
Computation and Communication - Two sides of one tapestry
Mon, Sep 28, 2009 @ 10:00 AM - 11:00 AM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
Speaker: Michael Gastpar,
University of California at BerkeleyAbstract: Networks have been studied in depth for the past several decades, but one feature has received little attention until recently: Interference. There is, of course, a good reason for this: In classical networks such as supply chains and the wired internet, interference can be addressed in a near-optimal fashion via simple protocols that avoid it. However, in the networks of prime interest today, such as wireless ad hoc networks, interference is often the dominant bottleneck and simply avoiding it entails major performance penalties. Therefore, the next important step is a thorough understanding of the nature of interference.In this talk, we argue that interference can be understood as computation: Multiple input signals are garbled together to produce a certain output. This is nothing but a certain computation performed on the input signals, possibly subject to noise or other stochastic effects. We show how this perspective inspires novel paradigms for thinking about communication in networks, including cooperation, "wireless network coding," and interference management. In particular, the computational perspective may help resolve the nagging question concerning the nature of information in networks: We have argued earlier that the "bit", a universal currency of information in single noisy channels, is inappropriate in general networks. A more appropriate currency of information could result from computational primitives, retaining algebraic structure as a fundamental property of information.Joint work with Bobak Nazer and, in part, with Jiening Zhan.Biography: Michael Gastpar (Ph.D. EPFL, 2002, M.S. UIUC, 1999, Dipl. El-Ing, ETH, 1997) is currently an Associate Professor in the Department of Electrical Engineering and Computer Sciences at the University of California, Berkeley. He was a visiting professor (2008/09) at the Technical University of Delft, The Netherlands. He was also a student in electrical engineering and philosophy at the Universities of Edinburgh and Lausanne, and a summer researcher in the Mathematics of Communications Department at Bell Labs, Lucent Technologies. His research interests are in network information theory and related coding and signal processing techniques, with applications to sensor networks and neuroscience. He won the 2002 EPFL Best Thesis Award, an NSF CAREER award in 2004, and an Okawa Foundation Research Grant in 2008.Host: Gerhard Kramer, gkramer@usc.edu, EEB 536, x07229Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248
Audiences: Everyone Is Invited
Contact: Gerrielyn Ramos
This event is open to all eligible individuals. USC Viterbi operates all of its activities consistent with the University's Notice of Non-Discrimination. Eligibility is not determined based on race, sex, ethnicity, sexual orientation, or any other prohibited factor.
