2012 talks

• Computational Social Science: Large-scale Measurement of Human Behavior
  Speaker: Sharad Goel, Yahoo! Researcher
  Date:Feb.6th Monday 4:00-5:30PM
  Venue:501, Bharti Building
  
  Abstract: With the increasing availability of network and behavioral data---from what we buy, to where we travel, to whom we know---we are now able to observe and quantify social processes to a degree that would have seemed impossible just a decade ago. These new microscopes into human activity not only have substantive implications for the social sciences, including economics, sociology, and psychology, but also raise challenging computational questions in large-scale data analysis. In this talk I will discuss several illustrative examples from this emerging discipline of computational social science, including the so-called viral diffusion of information, and the dynamics of long tail markets.
  
  

2010 talks

• Massive Data Computing
  Speaker: Dr. Pradeep K. Dubey, Intel Labs
  Date: Mar 11, 2010, at 4 pm
  
  Abstract: Computing platforms and the World Wide Web are undergoing architectural transitions driven by an unprecedented convergence of the need to process massive amounts of data with the availability of massive amounts of compute resources. This has algorithmic implications for traditional approaches to many common computational problems in visual computing and analytics. These applications are further distinguished by their service-oriented focus and real-time nature, aimed at delivering a significantly more immersive end-user experience. In this talk, we will discuss the architectural implications of these emerging throughput computing applications. We will also summarize the role we envision for Intels Throughput Computing Lab in this exciting field of research.
  
  Bio: Pradeep Dubey is a Senior Principal Engineer and Director of Throughput Computing Lab, part of Microprocessor and Programming Research division of Intel Labs. His research focus is computer architectures to efficiently handle new application paradigms for the future computing environment. He is currently responsible for throughput computing applications on Intels Larrabee architecture. Dubey previously worked at IBM's T.J. Watson Research Center, and Broadcom Corporation. He was one of the principal architects of the AltiVec multimedia extension to Power PC architecture. He also worked on the design, architecture, and performance issues of various microprocessors, including Intel i386, i486, and Pentium processors. He holds 26 patents and has published extensively. Dr. Dubey received his PhD in electrical engineering from Purdue University. He is a Fellow of IEEE.
  
  
• The past, present and future of Twitter
  Speakers: Pankaj Gupta and Jinen Kamdar, Twitter
  Date: Feb 26, 2010, at 4 pm
  
  Abstract: Learn about the history of Twitter and where the company is headed first-hand. We'd like to talk about the values the company was founded upon, as well as some of the transformative milestones that have shaped Twitter to make it the company that is today. Lastly, we'd like to talk about the plans for Twitter in India and the importance of the developer ecosystem and you.
  
  Bio: Pankaj Gupta works as a Sr. Researcher at Twitter , also helping drive strategy for Twitter in India. He did his PhD from Stanford University and B.Tech from IIT Delhi, both in Computer Science.
  
  Jinen Kamdar is the Product Manager for the Internationalization team at Twitter. He works with the team to help define products and features that will create an engaging, local Twitter experience across the globe.
  
  
• Horn-Ok-Please
  Speaker:Prof. Bhaskaran Raman, IIT Bombay
  Date: Feb 23, 2010, at 10 am
  
  Abstract:Road congestion is a common problem worldwide. Intelligent Transport Systems (ITS), in developed countries, seek to alleviate this problem using technology. But such ITS are mostly inapplicable in developing regions due to high cost and assumptions of orderly traffic. Efforts in developing regions have been few. We seek to develop low-cost ITS techniques to detect road congestion which will work in disorderly road conditions. We propose techniques based on road noise, especially vehicular honks, a characteristic feature of Indian roads. We take Indian traffic conditions as an example for our analysis but we believe that most of our claims and experimental results can be extended to other developing countries too.
  Our system comprises a pair of road side acoustic sensors, separated by a known distance. If a vehicle honks between the two sensors, its speed can be estimated from the Doppler shift of the honk frequency. In this context, we have developed algorithms for honk detection, honk matching across sensors, and speed estimation.
  We have done extensive experiments in semi-controlled settings as well as three different real road scenarios under different traffic conditions. From the road-side recordings, we have identified five possible metrics to characterize traffic state: 70th percentile of vehicle speed, percentile of speeds below 10 Kmph, number of honks, duration of honks and noise level. Statistical divergence of congested and free flowing traffic data, based on these five metrics, is verified at 99% confidence using the two sample KS and MWU tests. We have designed a threshold based technique to classify traffic state into congested and free flowing. n-fold cross validation of this technique gives minimum classification accuracy of 75% for all five metrics.
  
  Bio:Bhaskaran Raman received his B.Tech in Computer Science and Engineering from Indian Institute of Technology, Madras in May 1997. He received his M.S. and Ph.D. in Computer Science from University of California, Berkeley, in 1999 and 2002 respectively. He was a faculty in the CSE department at Indian Institute of Technology, Kanpur (India) from June 2003. Since July 2007, he is a faculty at the CSE department at Indian Institute of Technology, Bombay (India). His research interests and expertise are in communication networks, wireless/mobile networks, large-scale Internet-based systems, and Internet middleware services. His current focus and specific interest is in appropriate technology in the domain of communication and computing for rural use.
  
  
• Brain as a Dynamical System: Novel Approaches to Understanding Brain Function
  Speaker:Dr Rahul Garg, IBM Watson
  Date: Jan 21, 2010, at 3 pm
  
  Abstract:Advances in MRI imaging technologies have made it possible to observe the brain function in a non-invasive manner using the BOLD (blood-oxygenation level dependent) response. Functional MRI (fMRI) is a technique that measures that changes in the blood oxygenation level which is believed to be correlated with the neuronal activity. Despite these advances in technologies, the methods to analyze the fMRI data have been underdeveloped, relying primarily on the correlation of the observed response with the experimental protocol. However, some recent developments in analysis techniques based on machine learning methods have enabled newer experimental paradigms. For example, using functional connectivity analysis it is possible to analyze the interpret data from experiments involving resting state, virtual reality games, watching movies, listening to music etc. Several machine learning approaches have been used to predict behavior using fMRI data. These techniques are able to extract complex patterns of brain activity from fMRI data and provide additional scientific insights.
  In this talk I will describe our machine-learning based techniques including the Granger causality analysis that may be used to model brain as a dynamic system. These techniques are based on sparse regression and multivariate autoregressive modeling. We show that it is possible to extract additional information about brain function which is not accessible using the standard analysis techniques.
  
  
• AURA: Automated User-Centered Reasoning and Acquisition System
  Speaker: Dr Vinay Chaudhri, Program Director, AI Center, Stanford Research International, California
  Date: Jan 21, 2010 at 12:00 noon
  
  Abstract:AURA is an AI-motivated system being developed at Stanford Research International. The goal of this project is to build a generic knowledge acquisition capability to help scientists formulate their knowledge in Physics, Chemistry, and Biology. The short term goal is to Capture Knowledge and Answering Questions on Science Textbooks using concept graphs and reasoning engine. In this talk we shall discuss the design of AURA, and experimental evaluation of its ability to meet its goal. The knowledge formulation is done using directed graphs for structured objects and logical rules and equations for mathematical expressions. Posing questions to a knowledge-based system is another challenging task in Artificial Intelligence. The talk will briefly mention the efforts of SRI in this direction. The talk will also cover potential applications in bio-medical research.
  
  Bio:Dr Vinay Chaudhri is Program Director, Artificial Intellilgence Center, Stanford Research International, California. He did his Masters from IIT Kanpur, and Phd from University of Toronto. His research focuses on the science and engineering of large knowledge base systems and spans knowledge representation and reasoning, deductive question answering, knowledge acquisition, and innovative applications. He has worked with a wide range of knowledge representation and reasoning systems including LOOM, Ontolingua, Cyc, KM, and F-Logic. He was part of the team that developed the Open Knowledge Base Connectivity protocol for accessing knowledge bases. Since 2000 he has been working on a system to capture knowledge from science textbooks, and to answer questions at the back of the book.
  
  
• Stream Data Processing: For Pervasive Computing and Monitoring Applications
  Speaker:Prof. Sharma Chakravarthy, The University of Texas
  Date: Friday, January 15, 2010 at 8am
  
  Abstract:Stream data processing and complex event processing (CEP) provide a new approach for data fusion, pervasive computing, and monitoring, in general. Stream data processing poses many challenges. Two important characteristics of stream data processing - bursty arrival rates and the need for near real-time performance requirement - challenge the allocation of limited resources in the system. Operator/Query modeling, scheduling, and load shedding are major issues that differ from traditional database managements systems (DBMS). In this presentation, we discuss the above issues and solutions. We elaborate on the architecture of stream data processing and its synergistic integration with complex event processing. We briefly discuss modeling, scheduling, and load shedding aspects of stream data processing. We will conclude the talk with an architecture for synergistically integrating stream and complex event processing.
  
  
• Multi-armed bandits in sponsored search auctions
  Speaker:Yogeshwer Sharma, Cornell University
  Date: Monday, Jan 11, 2010 at 4:00 PM
  
  Abstract:In this talk, we are going to focus on online decision problems. In an online decision problem, the algorithm is provided with a set of alternatives, and selects one alternative in each of the $T$ sequential trials, deriving a reward for each selection. After $T$ trials, the total reward of the algorithm is compared with the best ``single-arm'' strategy which has the maximum reward in hindsight. The difference between the reward of the best single-arm strategy and that of the algorithm is called the regret, and one seeks algorithms whose regret is sublinear in $T$ and running time is polynomial in the problem size.
  We study an important class of online decision problems, called the multi-armed bandit problems, and extend the basic model in two important ways. In one extension, we model sponsored search auctions as a multi-armed bandit problem, and allow the alternatives (or advertisers in this case) to be strategic which can report possibly wrong rewards. We seek to provide incentives to advertisers so as to get socially efficient solutions. We prove that any socially efficient solution that provides right incentives to advertisers must suffer much higher regret than the regret suffered by algorithms for multi-armed problem without incentive issues. In the second extension, also motivated by sponsored search and resource selection in distributed systems, we allow the set of available alternatives to vary over time, provide a natural way to define the regret, and give policies that suffer low regret. We also prove that the regret suffered by our policies is information-theoretically the lowest possible.
  The talk is going to be self-contained and I will not assume any prior knowledge of online learning or sponsored search auctions. This talk is based on joint works with Moshe Babaioff (Microsoft Research), Robert Kleinberg (Cornell University), Alexandru Niculescu-Mizil (IBM Watson), and Alex Slivkins (Microsoft Research).
  
  

2009 talks

• Making Networks Programmable
  Speaker: Puneet Sharma, HP Labs
  Date: MONDAY, Dec 21, 2009, 11:30am.
  
  Abstract: The talk will present an overview of the networking research at HPLabs. Most vendors like Cisco, HP and Juniper have launched ``open'' platform designs for routers and switches, allowing code from customers or third-party vendors to run on their proprietary hardware. An open platform needs a programming interface, to provide switchlets sufficient access to platform features without exposing too much detail. I will discuss the design of an abstraction layer and API to support portability between vendor platforms, isolation between switchlets and both the platform and other switchlets, high performance, and programming simplicity. Also give some examples of in the areas of network power management and in-network processing that can be leverage network programmability of network comprising of such open platforms.
  
  Bio: Puneet Sharma is a Senior Research Scientist at HP Labs where he conducts research on Network Measurement and Monitoring, Wireless Networks, Quality of Service, and Overlay networks. Prior to joining the HP Labs in 1998, he received his PhD in Computer Science from the University of Southern California. He also holds a B.Tech. in Computer Science and Engineering from Indian Institute of Technology (IIT) Delhi. His work on Mobile Collaborative Communities was featured in the New Scientist Magazine. Currently his research focuses on building programmable networks and QoS mechanisms for fabric convergence. He has also participated in various standardization efforts. Recently he contributed to the UPnPs QoS Working Group efforts as co-author for QoSv3 standards. Earlier, he had co-authored the IETF standards RFCs on multicast routing protocol PIM. Puneet was the Co-Chair of 13th IEEE LANMAN Workshop in April 2004.
  
  
• Body-Wearable Sensor Devices for Health-care and Sports Monitoring
  Speaker: Prof. Vijay Sivaraman, University of New South Wales in Sydney, Australia
  Date: THURSDAY, Dec 17, 2009, 12 noon.
  Location: Room 510, Deptt. of Computer Sc. and Engg., Bharati School Building, IIT
  
  Abstract: This talk summarises our recent contributions related to body-wearable sensor devices used for physiological monitoring. In the context of healthcare applications, we empirically profile the impact of the body and human activity on the wireless link, and identify opportunities for energy savings by adapting radio transmit power. For sports applications, we profile the inter-connectivity on the soccer field, and develop models and algorithms for extracting player vital signs in real-time. We hope to foster discussion on new research challenges related to body-worn devices.
  
  Bio: Vijay Sivaraman received his B. Tech. from the Indian Institute of Technology in Delhi, India, in 1994, his M.S. from North Carolina State University in 1996, and his Ph.D. from the University of California at Los Angeles in 2000. His PhD studies were funded by a Fellowship from Bell Labs. He has worked at a silicon valley start-up manufacturing optical switch-routers, and is now an academic at the University of New South Wales in Sydney, Australia. His research interests include Optical Networking, packet switching and routing, and more recently wireless sensor networks for healthcare applications.