SUNMONTUEWEDTHUFRISAT
Conferences, Lectures, & Seminars
Events for August
-
AI SEMINAR-Fast large-scale optimization by unifying stochastic gradient and quasi-Newton methods
Fri, Aug 01, 2014 @ 11:00 AM - 12:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Jascha Sohl-Dickstein, Kahn Academy, Stanford University
Talk Title: Fast large-scale optimization by unifying stochastic gradient and quasi-Newton methods
Series: AISeminar
Abstract: Abstract:
I will present an algorithm for performing minibatch optimization that combines the computational efficiency of stochastic gradient descent (SGD) with the second order curvature information leveraged by quasi-Newton methods. These approaches are unified by maintaining an independent Hessian approximation for each minibatch. Each update step requires only a single minibatch evaluation (as in SGD), and each step is scaled using an approximate inverse Hessian and little to no adjustment of hyperparameters is required (as is typical for quasi-Newton methods). This algorithm is made tractable in memory and computational cost even for high dimensional optimization problems by storing and manipulating the quadratic approximations for each minibatch in a shared, time evolving, low dimensional subspace. Experimental results demonstrate improved convergence on seven diverse optimization problems. The algorithm is released as open source Python and MATLAB packages.
Optimizer available at:
https://github.com/Sohl-Dickstein/Sum-of-Functions-Optimizer
Paper reference:
Jascha Sohl-Dickstein, Ben Poole, and Surya Ganguli
Fast large-scale optimization by unifying stochastic gradient and quasi-Newton methods
International Conference on Machine Learning (2014)
http://arxiv.org/abs/1311.2115
Biography: Bio:
Jascha Sohl-Dickstein is an Academic Resident at the Khan Academy, and a visiting scholar in applied physics in Surya Ganguli's lab at Stanford University. He earned his PhD in 2012 in the Redwood Center for Theoretical Neuroscience at UC Berkeley, in Bruno Olshausen's lab. His research interests involve applying ideas from statistical physics and dynamical systems to problems in machine learning and neuroscience.
Host: Greg Ver Steeg
Webcast: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=925b9f53d4eb4964a37af20bacde2ad31dLocation: Information Science Institute (ISI) - 1135
WebCast Link: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=925b9f53d4eb4964a37af20bacde2ad31d
Audiences: Everyone Is Invited
Contact: Alma Nava / Information Sciences Institute
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. -
AI Seminar
Fri, Aug 08, 2014 @ 11:00 AM - 12:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Joel Tropp, Caltech
Talk Title: Finding Structure with Randomness: Stochastic Algorithms for Numerical Linear Algebra
Abstract: Computer scientists have long known that randomness can be used to improve the performance of algorithms. A familiar application is the process of dimension reduction, in which a random map transports data from a high-dimensional space to a lower-dimensional space while approximately preserving some geometric properties. By operating with the compact representation of the data, it is possible to produce approximate solutions to certain large problems very efficiently.
Recently, it has been observed that dimension reduction has powerful applications in numerical linear algebra and numerical analysis. This tutorial will offer a high-level introduction to randomized methods for some of the core problems in this field. In particular, it will cover techniques for constructing standard matrix factorizations, such as the truncated singular value decomposition and the Nystrom approximation. In practice, the algorithms are so effective that they compete withâ or even outperformâ classical algorithms. These methods are likely to have significant applications in modern large-scale learning systems.
Biography: Joel A. Tropp is Professor of Applied & Computational Mathematics at the California Institute of Technology. He earned his PhD degree in Computational Applied Mathematics from the University of Texas at Austin in 2004. Dr. Troppâs work lies at the interface of applied mathematics, electrical engineering, computer science, and statistics. This research concerns the theoretical and computational aspects of data analysis, sparse modeling, randomized linear algebra, and random matrix theory. Dr. Tropp has received several major awards for young researchers, including the 2007 ONR Young Investigator Award and the 2008 Presidential Early Career Award for Scientists and Engineers. He is also the winner of the 6th Vasil A. Popov prize and the 2011 Monroe H. Martin prize.
Host: Greg Ver Steeg
Webcast: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=20188709a05e4b678dfa2c2d588408ad1dLocation: Information Science Institute (ISI) - 11th floor large conference room
WebCast Link: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=20188709a05e4b678dfa2c2d588408ad1d
Audiences: Everyone Is Invited
Contact: Kary Lau
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. -
NL Seminar: "Rapid Generation of Pronunciation Dictionaries for New Domains and Languages
Fri, Aug 08, 2014 @ 03:00 PM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Tim Schlippe, Karlsruhe Institute of Technology
Talk Title: NL Seminar
Series: Natural Language Seminar
Abstract: Automatic speech recognition systems exist only for a small fraction of the more than 7,100 languages in the world since the development of such systems is usually expensive and time-consuming. Therefore, porting speech technology rapidly to new languages with little effort and cost is an important part of research and development. Pronunciation dictionaries are a central component for both automatic speech recognition and speech synthesis. They provide the mapping from the orthographic form of a word to its pronunciation, typically expressed as a sequence of phonemes. I will present innovative strategies and methods for the rapid generation of pronunciation dictionaries for new application domains and languages. Depending on various conditions, solutions are developed and proposed - starting from the simple scenario in which the target language can be found in written form on the Internet and we have a simple mapping between speech and written language - up to
the difficult scenario in which no written form for the target language exists. We embedded many of the tools implemented in this work in the Rapid Language Adaptation Toolkit. Its web interface is publicly accessible and allows people to build first speech recognition systems with little technical background.
Biography: Since 2008 Tim Schlippe is a research assistant and PhD student at Karlsruhe Institute of Technology (KIT), Institute for Anthropomatics, in Germany. At KIT he is involved in teaching and several projects. He has published multiple publications in the field of multilingual speech recognition. For his master's thesis he was as a visiting researcher at Carnegie Mellon University, doing research in the field of statistical machine translation. Tim Schlippe will finish his PhD in November 2014. His current research interests are: Multilingual speech recognition with a focus on rapid adaptation of speech recognition systems to new domains and languages, pronunciation modeling, and language modeling.
Home Page:
http://csl.ira.uka.de/~schlippe/
Host: Aliya Deri and Kevin Knight
More Info: http://nlg.isi.edu/nl-seminar/
Location: Information Science Institute (ISI) - 11th Floor Conf Rm (#1135)
Audiences: Everyone Is Invited
Contact: Peter Zama
Event Link: http://nlg.isi.edu/nl-seminar/
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. -
AI Seminar-Using Friends as Sensors to Detect Global-Scale Contagious Outbreaks
Fri, Aug 15, 2014 @ 11:00 AM - 12:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Esteban Moro, Universidad Carlos III de Madrid
Talk Title: Using Friends as Sensors to Detect Global-Scale Contagious Outbreaks
Series: Artificial Intelligence Seminar
Abstract: Recent research has focused on the monitoring of global-scale online data for improved detection of epidemics, mood patterns, movements in the stock market, political revolutions, box-office revenues, consumer behaviour and many other important phenomena. However, privacy considerations and the sheer scale of data available online are quickly making global monitoring infeasible, and existing methods do not take full advantage of local network structure to identify key nodes for monitoring. Here, we develop a model of the contagious spread of information in a global-scale, publicly-articulated social network and show that a simple method can yield not just early detection, but advance warning of contagious outbreaks. In this method, we randomly choose a small fraction of nodes in the network and then we randomly choose a "friend" of each node to include in a group for local monitoring. Using six months of data from most of the full Twittersphere, we show that this friend group is more central in the network and it helps us to detect viral outbreaks of the use of novel hashtags about 7 days earlier than we could with an equal-sized randomly chosen group. Moreover, the method actually works better than expected due to network structure alone because highly central actors are both more active and exhibit increased diversity in the information they transmit to others. These results suggest that local monitoring is not just more efficient, it is more effective, and it is possible that other contagious processes in global-scale networks may be similarly monitored. Finally, we show the effectiveness of the method in the recent Twitter activity during hurricane Sandy.
Biography: BSc in Physics (1994) from the University of Salamanca and PhD in Physics from the University Carlos III of Madrid (1999). Researcher at the University of Oxford (1999-2001) and Ramón y Cajal University Carlos III in Madrid (2003-2007). Associate Professor at Universidad Carlos III. He has published over 40 articles and have led and participated in over 20 projects funded by government and companies. His areas of interest are random processes, financial mathematics, viral marketing, social networks. He works as a consultant in social networks for the Institute of Knowledge Engineering. It was awarded "Shared University Award" from IBM in 2007 for modeling the spread of information in social networks and application to viral marketing. And Research Excellence Award in 2013 by the Carlos III University of Madrid.
Home Page/ Blog:
http://estebanmoro.org
markov.uc3m.es
Host: Kristina Lerman
Webcast: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=052c65d179f94ec995d9b00659196b761dLocation: Information Science Institute (ISI) - 11th Flr Conf Rm # 1135, Marina Del Rey
WebCast Link: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=052c65d179f94ec995d9b00659196b761d
Audiences: Everyone Is Invited
Contact: Peter Zamar
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. -
AI Seminar- A New Path Towards Machine Intelligence
Fri, Aug 22, 2014 @ 03:00 PM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Chris Adami , (Michigan State University)- Professor of Microbiology and Molecular Genetics & Professor of Physics & Astronomy
Talk Title: âA New Path Towards Machine Intelligence
Series: Artificial Intelligence Seminar
Abstract: For over fifty years, engineers have attempted to achieve machine intelligence that rivals human performance, but with only limited success in some specialized arenas such as chess. I will discuss what I believe is the central reason behind this failure, and how using the biological process of evolution can overcome that problem. I then discuss several applications of our "evolutionary intelligence" approach to understand brains and behavior.
Biography: Dr. Adami is Professor for Microbiology and Molecular Genetics & Physics and Astronomy at Michigan State University in East Lansing, Michigan. As a computational biologist, Dr. Adamiâs main focus is Darwinian evolution, which he studies theoretically, experimentally, and computationally, at different levels of organization (from simple molecules to brains). He has pioneered the application of methods from information theory to the study of evolution, and designed the âAvidaâ system that launched the use of digital life (mutating and adapting computer viruses living in a controlled computer environment) as a tool for investigating basic questions in evolutionary biology. He was also a Principal Scientist at the Jet Propulsion Laboratory where he conducted research into the foundations of quantum mechanics and quantum information theory. Dr. Adami earned a BS in physics and mathematics and a Diplom in theoretical physics from the University of Bonn (Germany) and MA and PhD degrees in physics from the State University of New York at Stony Brook. He wrote the textbook âIntroduction to Artificial Lifeâ (Springer, 1998) and is the recipient of NASAâs Exceptional Achievement Medal. He was elected a Fellow of the American Association for the Advancement of Science (AAAS) in 2011.
Host: Greg Ver Steeg
Webcast: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=c95288c6083d483e9767155c41be459a1dLocation: Information Science Institute (ISI) - 11th Flr Conf Rm # 1135, Marina Del Rey
WebCast Link: http://webcasterms1.isi.edu/mediasite/Viewer/?peid=c95288c6083d483e9767155c41be459a1d
Audiences: Everyone Is Invited
Contact: Peter Zamar
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. -
NL Seminar- [Intern talk] Determinental Point Processes for Human-Augmented Machine Translation
Fri, Aug 22, 2014 @ 03:00 PM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Allen Schmaltz, Harvard University
Talk Title: Determinental Point Processes for Human-Augmented Machine Translation
Series: Natural Language Seminar
Abstract: This talk will introduce languageFractal, an online system for human-augmented machine translation (MT) that aims to incorporate monolingual speakers into the translation pipeline in a cost-effective manner. The essential principle is to take a middle ground between pure MT and a fully crowdsourced approach by augmenting MT results with human corrections in an iterative cycle. To efficiently emit phrases and sentences to users and to effectively explore the space of possible translation options, we propose the use of determinantal point processes (DPPs), which can be used to model subset selection problems in which diversity of the subset is a desirable characteristic.
I will provide a brief tutorial on DPPs (including L-ensembles and the structured variant), and I will present an overview of our formulation of DPPs for dynamic programming problems in the context of the human-augmented machine translation pipeline. I will also introduce the languageFractal pilot and pipeline, the full trials of which will run through the 2014-2015 academic year at Harvard University.
Biography: Allen Schmaltz is a Ph.D. student in Computer Science in the School of Engineering and Applied Sciences at Harvard University (2013-present; S.M. 2014), working with Stuart Shieber. He is interested in formal, statistical, and human-augmented machine learning approaches for computational linguistics. Before starting his Ph.D. in Computer Science, he completed the better part of an additional Ph.D. in the (quantitative) social sciences at Harvard University (2010-2013), received a M.A. from Stanford University (2010), and received a B.A. from Northwestern University (2006). Earlier in his academic career he also studied at Cornell University and in Yokohama, Japan, among other places.
Host: Aliya Deri and Kevin Knight
More Info: http://nlg.isi.edu/nl-seminar/
Location: Information Science Institute (ISI) - 11th Flr Conf Rm # 1135, Marina Del Rey
Audiences: Everyone Is Invited
Contact: Peter Zamar
Event Link: http://nlg.isi.edu/nl-seminar/
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. -
NL Seminar- Toward Semantic Parsing [Intern final talk]
Fri, Aug 29, 2014 @ 03:00 PM - 04:00 PM
Information Sciences Institute
Conferences, Lectures, & Seminars
Speaker: Allen Schmaltz and Julian Schamper, Harvard University and RWTH Aachen University.
Talk Title: Toward Semantic Parsing [Intern final talk]
Series: Natural Language Seminar
Abstract: Semantic parsing has potential applications in a number of areas, including machine translation and machine reading, among many others. In this talk we will present our initial work on the parsing task for the semantic representation language known as Abstract Meaning Representation (AMR). The task is to take an English sentence and transform it into its semantic representation.
We will present a series of approaches and associated results, providing guidelines for future work in this area. We will show approaches using heuristics, tree transducers, and probabilistic context free grammars. We will also present approaches for AMR rule extraction for the applicable formalisms. In doing so, we will also highlight challenges relative to syntactic parsing.
Additionally, we will provide a map for the future directions in AMR parsing that we plan to pursue in the fall.
Biography: Allen Schmaltz is a Ph.D. student in Computer Science in the School of Engineering and Applied Sciences at Harvard University (2013-present; S.M. 2014), working with Stuart Shieber. He is interested in formal, statistical, and human-augmented machine learning approaches for computational linguistics. Before starting his Ph.D. in Computer Science, he completed the better part of an additional Ph.D. in the (quantitative) social sciences at Harvard University (2010-2013), received a M.A. from Stanford University (2010), and received a B.A. from Northwestern University (2006). Earlier in his academic career he also studied at Cornell University and in Yokohama, Japan, among other places.
Julian Schamper studies computer science at RWTH Aachen University. He did his bachelor thesis in the field of deciphering foreign language and works as a student research assistant at Prof. Hermann Ney's Human Language Technology and Pattern Recognition Group.
Host: Aliya Deri and Kevin Knight
More Info: http://nlg.isi.edu/nl-seminar/
Location: Information Science Institute (ISI) - 11th Flr Conf Rm # 1135, Marina Del Rey
Audiences: Everyone Is Invited
Contact: Peter Zamar
Event Link: http://nlg.isi.edu/nl-seminar/
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.
