National Conference on Communications - NCC 2011, IISc Bangalore

28-30 January 2011

Indian Institute of Science

Bangalore

Tutorials

All tutorials will be on Friday, 28 January 2011.

Morning tutorials: 09:00 - 12:20
1.Onkar Dabeer, TIFR Mumbai
Collaborative filtering Abstract

Recommendation systems are commonly used in e-commerce to suggest relevant content to users. This tutorial considers the related problem of estimating missing user-item ratings based on the available ratings. While a number of algorithms have been proposed for this purpose, the dominant focus has been on empirical performance evaluation over massive datasets. However, over the past two years, several researchers have proposed suitable mathematical models for this problem, studied fundamental limits, and shown provably good algorithms for this problem. This tutorial covers some of this recent work. Broadly speaking, there are two dominant models for the rating data: 1) a low-rank matrix model, and 2) a rearranged block constant matrix with erasures and noise. We will briefly cover algorithms for completion of low-rank matrices, which are inspired by compressed sensing. For the second model, we will present fundamental limits for the recovery of the entire matrix. Moreover, we will analyze collaborative filters based on "popularity amongst friends" under the second model. Several simulation results on datasets such as Movielens and Netflix will also be discussed in light of the theoretical results. We will conclude by suggesting open directions for future work, including extension to other kinds of estimation problems.

Biography

Onkar Dabeer got his B.Tech and M.Tech from IIT Bombay in 1996 and 1998 respectively. He got his PhD from the University of California, San Diego, in 2002. After a postdoc at UC Santa Barbara and a year at Qualcomm Inc, he joined the faculty if the School of Technology and Computer Science, Tata Institute of Fundamental Research, Mumbai in August 2004. At TIFR he pursues his interests in estimation theory (with current focus on problems arising in e-commerce/internet/social network related datasets), multi-gigabit wireless networks, and information theory.PDF of presentation
2.Jishnu A., Tejas Networks
Backhaul transport network Abstract

The backhaul transport network consists of optical/microwave networks which transports the bits/bytes from various last mile access technologies like DSLAM (Copper), BTS (Wireless), GPON (Fiber) to the core IP/MPLS network consisting of BRAS, BSC/MSC/RNC. The tutorial would cover the challenges in the transport network like protection, scalability, synchronization etc and how each of them are addressed. One of the most important characteristics of transport network is the requirement for sub 50 ms resiliency. The tutorial would explain how 50ms protection is obtained using various topologies like ITU-T G.8031 ELPS, G.8032 ERPS. As the backhaul transport network evolves from a TDM to a more packet based network, there is a requirement to handle network level scalability. There are two approaches to scalability based on MPLS-TP and PBB/PBB-TE. The need for smooth handover in TDD based wireless last mile technologies require uS level phase synchronization accuracies which are provided using SyncE (ITU-T G.8261) and PTP (IEEE 1588v2). The talk would cover all this aspects as well.

Biography

Jishnu A. got his B.Tech. degree in electronics and communication from NIT Calicut and M.Sc. (Engg.) from the Indian Institute of Science, Bangalore. He is currently a Lead Engineer at Tejas Networks India Limited, Bangalore.
3.Kaushik Majumdar, ISI Bangalore
Sanjib Sinha, NIMHANS Bangalore
Electroencephalogram (EEG) Abstract

Human scalp EEG is a real time electrical signature of the neuronal activities inside the brain. It is a biological signal with wide ranging implications in health care systems and cognitive science research. The opening one hour lecture will be devoted to very long duration continuously monitored EEG signals (called video EEG) of epileptic patients. The next two hour lecture will be devoted to one of the greatest challenges of the quantitative EEG (qEEG), known as the source localization problem. This is actually an ill-posed problem, but with immense practical value, and therefore plenty of research efforts are being devoted in this direction.

1) Clinical EEG: EEG-video monitoring is the highest level of epilepsy monitoring and the gold standard (S. R. Benbadis, p. 989 in Sleep: A Comprehensive Handbook, T. L. Lee-Chiong (ed.), Wiley, 2006). Dr. Sanjib Sinha of the Neurology Department in NIMHANS, Bangalore is an expert in this area who will be giving a lecture on the EEG-video monitoring of the epileptic patients. In the lecture it might be emphasized how the scalp EEG evolves along with the video recorded behavior of the patient during the prolonged recordings.

2) Source localization: EEG and MEG have the finest temporal resolutions among all the neural signals and hence most capable of representing the dynamic brain at the millisecond range. This advantage is however largely nullified by the poorest spatial resolution of the scalp EEG among all the neural signals. So, the cortical source localization of the scalp EEG is a very important research topic.(for excellent reviews see Hallez, et al., "Review on solving the forward problem in EEG source analysis," J. NeuroEng. Rehab. (open access), vol. 4, available at http://www.jneuroengrehab.com/content/4/1/46 , 2007 and Grech, et al., "Review on solving the inverse problem in EEG source analysis," J. NeuroEng. Rehab. (open access), vol. 5, available at http://www.jneuroengrehab.com/ content/5/1/25, 2008). Various quantitative methods of human scalp EEG source localization will be discussed. The two hour lecture will be divided into (a) forward problem and (b) inverse problem. Kaushik Majumdar of ISI, Bangalore Centre, will give this part of the tutorial.

a) Forward problem: To have a meaningful solution of the ill-posed problem of scalp EEG source localization a realistic head-model of the subject is built up. Head can be thought of as layers of tissues, where each layer has clearly defined boundaries. Electric field travels from source through the tissue layers to the scalp surface. This is modeled either by boundary element or by finite element method. This part of the talk will be centered mostly on the boundary element methods (BEM). Solution of the forward problem can be summarized in a matrix form, called EEG gain matrix, which maps the source space into the sensor (scalp electrode) space.

b) Inverse problem: The focus of the inverse problem is to find a map from sensor space back to the source space. Gain matrix is generally not a square matrix, so a straightforward linear inverse is not an option. However such a mapping is possible to obtain under various constraints. Particular emphasis will be laid on the most successful methodology so far, called the minimum norm inverse. A novel method to estimate the cortical sources by phase synchronization and signal power distribution among the scalp EEG channels will also be discussed.

Biography

Kaushik Majumdar is with the System Science and Informatics Unit of the Indian Statistical Institute, Bangalore Centre. Sanjib Sinha is with the Neurology Department of NIMHANS, Bangalore.PDF of presentation (Kaushik Majumdar)
PDF of presentation (Sanjib Sinha)
Afternoon tutorials: 13:30 - 16:50
4.Jayashree Ratnam, IIT Kharagpur
Fibre to the premises: a next-generation access network solution Abstract

Broadband access networks using optical fibers have emerged as the ultimate "last mile" wired solution for delivering services like HDTV, video-on-demand, video conferencing in addition to high-speed data services to the subscriber premises. Many incumbent local exchange carriers in North America, Asia-Pacific and Europe are gearing up for a migration to such optical access infrastructure. Passive architectures with point-to-multipoint type of connectivity, between the central office and customer premise terminations called passive optical networks (PONs) are currently the most popular fiber-to-the-premises solutions due to their cost-effectiveness. Single carrier PONs like BPON, EPON, GPON etc. have already been standardized by ITU-T and IEEE. However the futuristic needs of the access segment in telecommunication sector, can be effectively fulfilled only by incorporating wavelength-division multiplexing (WDM) in the network. Several research efforts and test-bed studies to examine the feasibility of such WDM-based PONs were reported in the recent past.

This tutorial traces the evolving scenario of access networks starting with time-division multiplexing-based PONs (TDMPONs) to the future-proof WDMPONs. A Brief treatment to the currently available single-wavelength PON standards (viz., BPON, EPON etc.) is given, highlighting their salient features as well as their limitations. The presentation then moves on to next generation access solutions viz., WDMPONs, to examine a few typical network architectures. Reports on some of the test-bed studies on WDMPONs and their relevance to a particular network setting are also discussed. Thereon the tutorial deliberates on the outcome of several research studies conducted on physical (device and system level) and network layer (medium access and scalability) issues of the PONs and their implications in designing a cost-effective network model. The tutorial gives adequate coverage to various optical network unit (ONU) configurations (collocated with user-premises) which realize wavelength-independent transmission at the ONU. In the last leg of the presentation, the tutorial touches upon issues related to transmission impairments, resource provisioning and delay-sensitive access protocols in WDM PONs, with the help of results obtained through analytical studies. Finally, the tutorial discusses the future directions in research and current status of standardization activity for next generation optical access solutions.

Biography

Jayashree Ratnam received the PhD and M. Tech. degrees from Indian Institute of Technology, Kharagpur, India in 2009 and 1997 respectively. She took the B. Tech degree in Electronics and Communication Engineering from Jawaharlal Nehru Technological University, Hyderabad, India in 1985. She served for about 12 years at IIT Kharagpur both as a research staff member and as a Scientific Officer. Presently she is working as a Research Consultant in Vodafone Essar IIT Center of Excellence in Telecommunications at IIT Kharagpur. She carried out investigations on transmission impairments and medium access control issues in WDM-Based Optical Access Networks for her PhD dissertation. She published her research results in four international journals and seven international/national conference proceedings. Her paper presented in IEEE ANTS 2008 at IIT Bombay received best paper award. Currently she is on the Editorial Board of "Far East Journal of Electronics and Communications (FJEC-ISSN:0973-7006)". Her biographical sketch has been selected for "Marquis Who's Who in the World 2010 edition". Her research interests include PHY and MAC layer issues in WDM-based optical access networks and Optical CDMA-based communication networks.PDF of presentation
5.Sibi Raj Bhaskaran Pillai, IIT Bombay
Compressed sensing and applications Abstract

Compressed sensing (CS) is a collection of techniques for sparse signal acquisition and reconstruction, in a resource efficient manner. Following the seminal works by Candes et al (IEEE IT Feb2006) and Donoho (IEEE IT Apr2006), CS has been a very active research area with a multitude of applications and new discoveries. This talk chronicles some of the major contributions to CS, and attempts a crisp explanation of the underlying mathematical engine. We will supplement the theory with several possible applications in communications, signal processing, computing, biomedical etc.

Biography

Sibi Raj B. Pillai is an assistant professor at IIT Bombay. Prior to joining IIT Bombay in 2009, he was a research fellow at the University of Melbourne, Australia.PDF of presentation
6.Shalabh Bhatnagar, IISc Bangalore
Reinforcement learning – an overview Abstract

Reinforcement learning (RL) is one of the most active research areas in machine learning (ML) and artificial intelligence (AI). While it has its roots in AI, RL has found tremendous applications in problems of stochastic control. The aim of this tutorial is to familiarize the audience with the techniques and algorithms used in RL as well as to present the theory behind these. Towards this end, we shall first briefly discuss the basic formulations and concepts from Markov decision processes (MDP). Next, we shall give an overview of stochastic approximation algorithms and discuss convergence issues. It turns out that most of the RL algorithms are in fact stochastic approximation algorithms that can be viewed as solving an associated stochastic control problem. We shall present RL algorithms under both scenarios – (a) full state and (b) feature-based state representations. Finally, a few interesting applications of RL will also be presented.

Biography

Shalabh Bhatnagar is an associate professor at the Indian Institute of Science, Bangalore.PDF of presentation

NCC Home

Important Dates

Paper submission begins:
22 June 2010
Paper submission deadline:
06 September 2010
17 September 2010
22 September 2010
27 September 2010 (final)
Acceptance notification:
15 November 2010
Final manuscript upload deadline:
08 December 2010

IISc

Financial Sponsors

Diamond

Platinum

Gold

Silver

Technical Co-sponsor

IEEE Communications Society

IEEE

IEEE Bangalore Section