Events

Faculty Colloquium (25 Nov. '16 )

Past Colloquiums:

 

Title Speaker Time & Venue Video (optional)
The Communication Complexity of Secret Key Generation

 Navin Kashyap,    ECE Dept., IISc

Date: Friday, 23 September 2016
Time: 4pm(Coffee/tea at 3.45pm)
Venue: Golden Jubilee Hall, ECE Department.

Link

Abstract:

We consider the problem of secret key generation within the multiterminal source model of Csiszar and Narayan. In this model, there are a finite number, m, of terminals, the i-th terminal having access to i.i.d. realizations of a discrete random variabel X_i. The joint distribution of (X_1,...,X_m) is known to everyone. The terminals communicate over a noiseless public channel, using which they must agree upon a secret key that is almost independent of the public communication. The secret key capacity, i.e., the largest rate of a secret key that can be generated by the m terminals, was determined by Csiszar and Narayan in 2004.
In this talk, we examine the communication complexity of secret key generation, which is defined to be the minimum sum rate of communication required to generate a secret key of a given rate. Equivalently, we address the question of what is the largest rate of a secret key that can be generated using a communication of total rate R. We provide an overview of what we know so far about this problem. In particular, we provide a complete and precise answer for the important special case of the pairwise independent network (PIN) model.
This is joint work with Chung Chan, Manuj Mukherjee and Qiaoqiao Zhou.

 

Speaker

Date

Manoj Varma

28 Oct 2016

Rajiv Soundararajan

25 Nov 2016

 

 

Title Speaker Time & Venue Video (optional)
If we can make computers play chess, why can't we make them see?

 S. P. Arun,    CNS/ECE Dept., IISc

Date: Friday, 26 August 2016
Time: 4pm(Coffee/tea at 3.45pm)
Venue: Golden Jubilee Hall, ECE Department.

Link

Abstract:

If we can make computers play chess and even Jeopardy, then why can't we make them see like us? This is a particularly perplexing question when we consider how easy we find the act of seeing which we perform countless times each day with virtually no errors. What makes vision such a hard problem? How does the brain accomplish vision? To answer these questions we perform behavioral tests of vision in humans as well as recordings from neurons in the visual cortex of monkeys. I will describe some of our recent findings elucidating object recognition at the behavioral and neuronal level.

 

Speaker

Date

Kausik Majumdar

27 May 2016

 

Title Speaker Time & Venue Video (optional)
Universal Multiparty Data Exchange

 Himanshu Tyagi ,    Dept. of ECE., IISc

Date: Friday, 6 May. 2016
Time: 4pm(Coffee/tea at 3.45pm)
Venue: Golden Jubilee Hall, ECE Department.

Link

Abstract:

We consider the problem of secret key generation within the multiterminal source model of Csiszar and Narayan. In this model, there are a finite number, m, of terminals, the i-th terminal having access to i.i.d. realizations of a discrete random variabel X_i. The joint distribution of (X_1,...,X_m) is known to everyone. The terminals communicate over a noiseless public channel, using which they must agree upon a secret key that is almost independent of the public communication. The secret key capacity, i.e., the largest rate of a secret key that can be generated by the m terminals, was determined by Csiszar and Narayan in 2004.
In this talk, we examine the communication complexity of secret key generation, which is defined to be the minimum sum rate of communication required to generate a secret key of a given rate. Equivalently, we address the question of what is the largest rate of a secret key that can be generated using a communication of total rate R. We provide an overview of what we know so far about this problem. In particular, we provide a complete and precise answer for the important special case of the pairwise independent network (PIN) model.
This is joint work with Chung Chan, Manuj Mukherjee and Qiaoqiao Zhou.

 

Title Speaker Time & Venue Video (optional)
Towards Smart Electromagnetic Solvers: Learning-based or Physics-based

 Dipanjan gope ,    Dept. of ECE., IISc

Date: Friday, 18 March. 2016
Time: 4pm(Coffee/tea at 3.45pm)
Venue: Golden Jubilee Hall, ECE Department.

Link

Abstract:

The solution of Maxwell's equations for irregular 3D structures have been used traditionally for oil-exploration, detection of radar-cross-section of military vehicles, antenna analysis, bio-medical applications and more recently in predicting electrical performance of radio-frequency (RF) and high speed circuits and systems. These solvers treat each model independently regardless of any similarity between previously solved models and therefore sacrifices the possibility of accelerating a model solution from knowledge harnessed from a prior solution of a similar model. This is a missed opportunity particularly in the solution of design variants which involve multiple models with near–identical geometrical features.

 

Title Speaker Time & Venue Video (optional)
Realizing the Power of MIMO Signal Processing

 Professor K.V.S. Hari,    Dept. of ECE., IISc

Date: Tuesday, 16 Feb. 2016
Time: 4pm(Coffee/tea at 5pm)
Venue: Faculty Hall, Main Building.

Link

Abstract:

Signal Processing systems, which use multiple inputs (signals/sequences) and process them using transformations to generate multiple outputs, are called Multiple-Input-Multiple-Output (MIMO) systems. MIMO systems describe several phenomena in fields spanning biology, neuroscience, chemical engineering, electrical engineering and others. MIMO Signal processing deals with the design of theoretical models, algorithms and implementation architectures. In this talk, we present several examples of MIMO systems and then consider a few applications like direction estimation, wireless communication and indoor localization. The impact of work carried out by members of the lab will be highlighted in the context of understanding new algorithms, contribution to industry standards and development of working prototypes.

 

Title Speaker Time & Venue Video (optional)
Fortune telling with IoT (Internet of Things)

 Prof. Bharadwaj Amrutur,        Dept. of ECE., IISc

Date: Friday, 27 Nov. 2015
Time: 4pm (Coffee/tea at 3.45pm)
Venue: Golden Jubilee Hall, ECE Dept.

Link

Abstract:

Who wouldn't pay to know the future? I bet anyone will  be glad to know that the brakes of their car will fail within six months and will promptly take it for service. Or that a Myocardial Infarction (heart attack) is imminent in the next seven days and better visit the hospital. Or maybe they will get mugged in the next 10 minutes - so better take some evasive action or call in the cops. Awareness of the current situation, context and  history, coupled with a model of the system - both self and environment along with a model of failure will enable forecasting of bad tidings.  We expect that IoT along with AI (Artifical Intelligence)   will enable such new services and we will explore these in greater detail in the talk.

 

Title Speaker Time & Venue Video (optional)
Visible Light Communication: An Emerging Area in Wireless

  Prof. A. Chockalingam,         Dept. of ECE., IISc

Date: Friday, 30 Oct. 2015
Time: 4pm (Coffee/tea at 3.45pm)
Venue: Golden Jubilee Hall, ECE Dept.

Link

Abstract:

Wireless communication using visible light wavelengths (400 to 700 nm) in indoor local area network environments is emerging as a promising area of research. Visible light communication (VLC) is evolving as an appealing complementary technology to radio frequency (RF) communication technology. In VLC, simple and inexpensive light emitting diodes (LED) and photo diodes (PD) act as signal transmitters and receptors, respectively, replacing more complex and expensive transmit/receive RF hardware and antennas in RF wireless communication systems. Other favorable features in VLC include availability of abundant visible light spectrum at no cost, no licensing/RF radiation issues, and inherent security in closed-room applications. The possibility of using the same LEDs to simultaneously provide both energy-efficient lighting as well as high-speed short-range communication is another attractive feature. In addition, the potential to use multiple LEDs and PDs in multiple-input multiple-output (MIMO) array configurations has enthused MIMO wireless researchers (including the speaker) to take special interest in VLC.  This talk will provide an overview of VLC and the associated technological challenges and opportunities. It will also highlight some of our recent contributions in the area of multiple-LED wireless communications.

 

Title Speaker Time & Venue Video (optional)
Noisy Index Coding and Side-Information

 B. Sundar Rajan,                             Dept. of ECE., IISc

Date: Friday, 18 Sep. 2015
Time: 4pm (Coffee/tea at 3.45pm)
Venue: Golden Jubilee Hall, ECE Dept.

Link

Abstract:

An index coding problem arises when a single source intends to communicate to a number of receivers over a noiseless broadcast channel. The sender has a number of messages and each receiver desires a specific subset of messages while having another subset of messages as side-information. A single encoding of messages can be transmitted in each channel use to all the receivers. The objective is to design an encoding scheme with minimum number of channel uses that satisfies all receivers - known as the noiseless index coding problem. For the binary messages this amounts to minimum bandwidth solution if only binary transmission is assumed -such solutions (not necessarily unique) are called optimal index codes. Noiseless Index Coding Problem not only remains open, but also has been shown to include as special cases a number of difficult problems in both wired and wireless network settings – such as the general multiple unicast problem with linear network coding, source coding with an informed source (source coding with side-information), multiway relay networks and the topological interference management problem in wireless networks to name a few.

In this talk we focus on noisy index coding problem where each transmission may be received with error by the receivers. In this context we show that different optimal index codes of the noiseless index coding problem perform differently in terms of probability of error. This motivates the problem of finding the number of optimal index codes for a noiseless index coding problem. We present a lower bound and exact value on the number of optimal index codes for different special cases. Also, we illustrate with few examples that side-information can lead to simultaneous power and bandwidth gain for some users and trade-off between power and bandwidth gain for some other users depending upon the amount of side-information and the choice of the index code used. [Joint work with the students: R. Kavitha, A. Chandramouli, Anoop Thomas and Anjana Mahesh.]

 

Title Speaker Time & Venue Video (optional)
Sequential Decision Making in Complex Environments

 Dr. Aditya Gopalan,                     Dept. of ECE., IISc

Date: Friday, 21 Aug. 2015
Time: 4pm (Coffee/tea at 3.45pm)
Venue: Golden Jubilee Hall, ECE Dept.

Link

Abstract:

Sequential decision making or online learning is concerned with studying how an agent can learn to perform a task with repeated actions and the ensuing feedback. An increasing number of modern-day automated systems are tasked with learning to make sequential decisions by utilizing evolving data, e.g., Internet recommendation engines, dynamic pricing algorithms, automated trading systems, etc.

We will discuss a widely studied model of decision making under uncertainty called the Multi-Armed Bandit, where a decision maker optimizes by repeatedly playing "arms" or actions that represent unknown payoff distributions and learning from past payoffs. A rich set of algorithmic and statistical problems emerges when the space of actions and payoffs is large and/or possesses complex structure, e.g., high-dimensional, linearly parameterized payoffs, or combinatorially large decision spaces as in network routing. We will explore such bandit variants and review well-known approaches to bandit optimization. We will also present recent progress in understanding the behaviour of a natural, Bayesian-inspired algorithm, dubbed Thompson (or posterior) sampling, that enjoys favourable empirical performance. In particular, we will present results that show that the
algorithm is highly flexible in optimizing in the presence of general, complex information structures and complex time dynamics as in Markov Decision Processes.

 

Title Speaker Time & Venue Video (optional)
Wideband adaptation and scheduling in OFDM-based 4G cellular systems

 Prof. Neelesh B. Mehta,                     Dept. of ECE., IISc

 Date31st July,  2015 (Friday)            Time: 4:00 pm (tea/coffee at 3:45pm).   Venue: Golden Jubilee Hall, ECE Dept.

Link

Abstract:

Contemporary cellular systems such as Long Term Evolution (LTE) are based on orthogonal frequency division multiplexing (OFDM). These systems divide the available bandwidth into orthogonal subchannels, and achieve high spectral efficiencies and data rates using techniques such as rate adaptation, multiple antennas, and scheduling. In these wideband systems, a packet is transmitted over a group of subchannels, which see different gains. Therefore, the choice of the user for a subchannel and the rate for a packet must be based on the vector of signal-to-noise-ratios (SNRs) seen by these subchannels. This requires us to revisit how adaptation and scheduling are implemented and analyzed in these systems.

We develop a new, comprehensive framework to analyze the cell throughput of such systems. Our framework accounts for wideband rate adaptation, scheduling, and feedback delays. It also captures the critical role played by the channel quality feedback scheme, which provides the base station with a limited portion of the large channel state information it needs to carry out adaptation and scheduling. Our framework makes use of a non-linear, physical layer abstraction technique called Exponential effective SNR mapping (EESM), and develops a new approximate statistical model for it. Unlike prior work, most of which is simulation-based, our work encompasses correlated and independent subcarriers, and various multiple antenna diversity modes. It is accurate over a wide range of delays.

 

Title Speaker Time & Venue Video (optional)
Enhancing information theoretic capacity of a communication channel Prof. Vinod Sharma, Dept. ECE., IISc Time: 4:00 pm (tea/coffee at 3:45pm). Date3rd July.  2015 (Friday). Venue: Golden Jubilee Hall, ECE Dept. Link

Abstract:

For a degraded wireless channel, the secrecy capacity is the capacity of the main channel minus the capacity of the evesdropper's channel. This leads to possibly a significant loss in the transmission rate, making researchers think about the possible ways to mitigate this loss. This talk will present our efforts in this direction.
First we will argue that taking secrecy criterion as probability of error of the message to the evesdropper can be sensible. We will mention several other recent studies using it. Then we will present simple consequences of this, not considered by others.
Next we show that within the framework of usual secrecy criterion of mutual information to the evesdropper, if we study the usual communication set up of transmitting multiple messages over time, using previously transmitted messages as keys, can allow transmission at the Shannon capacity.
This work is part of the PhD thesis of my student Shahid Mehraj Shah.

 

Title Speaker Time & Venue Video (optional)
Estimating the Capacity of the 2-D Hard Square Constraint Using Generalized Belief Propagation Prof. Navin Kashyap, Dept. ECE., IISc Time: 4:00 pm (tea/coffee at 3:45pm). Date24th Apr.  2015 (Friday). Venue: Golden Jubilee Hall, ECE Dept. Link

Abstract:

A binary mxn array is said to satisfy the hard square constraint if no row or column of the array contains ones in consecutive positions. Let N(m,n) denote the number of such arrays. The capacity of the hard square constraint is the limit, as m,n go to infinity, of the quantity (1/mn)*(log N(m,n)). Determining this capacity exactly is a long-standing open problem; its numerical value is known up to ten decimal places. Recently, Sabato and Molkaraie demonstrated via simulations that the capacity of the hard square constraint can be approximated extremely well using a generalized belief propagation (GBP) algorithm. They did not, however, give an analytical explanation for their observations. In this talk, we outline an approach for analyzing GBP with a view to giving guarantees for the accuracy of the GBP estimate of the capacity of the hard square constraint. More generally, our approach provides a means of analyzing the performance of GBP in estimating the partition function of any Ising model on the 2-D lattice.

This talk is based on ongoing work with Eric Chan, Sidharth Jaggi and Pascal Vontobel at the Chinese University of Hong Kong. It will be kept self-contained and accessible to a broad audience. 

 

Title Speaker Time & Venue Video (optional)
Micro-Opto-Electro-Mechanical Systems Prof. T. Srinivas, Dept. ECE., IISc Time: 4:00 pm (tea/coffee at 3:45pm). Date27th Mar. 2015 (Friday). Venue: Golden Jubilee Hall, ECE Dept. Link

Abstract:

Micro-opto-electro-mechanical system (MOEMS) technology promises compact and functional devices for applications ranging from optical communications to optical sensors.  In this presentation,  common types of MOEM sensors such as for pressure, vibration, rotation, acceleration and force are reviewed  along with their structure and performance. The focus here is on sensors based on light guidance in integrated optical waveguide devices controlled by MEMS structures.  Recently there is considerable interest in integrated optic ring resonators and photonic bandgap structures  as they provide wavelength and spectral based measurements, along with providing more compact structures.   Here we present our recent work on two types of MOEMS sensors. First  example is a novel optical MEMS accelerometer based on double ring resonator(DRR)  for vibration sensing application. The sensor is basically structured on two cascaded diaphragms with a small proof mass suspended in the middle of each diaphragm and serially coupled DRR is integrated along the edge of diaphragms to gain maximum stress on the resonators. A serially coupled DRR with different radii are investigated that, effective free spectral range (FSR) of the resonator is dramatically enhanced due to the vernier effect between the two rings. Thus, when acceleration is applied, it experience forces in each diaphragm which displace the masses, that induce identical and uniform stress in each ring, which brings a change in refractive index in the rings, leading to change in the output spectrum shift providing the measure of accelerometer. Sensitivity as high as 6.542pm/g has been achieved and it is capable of measuring within range of (0-4270g). As another example a force sensor configuration in which  a photonic crystal (PC) ring resonator is integrated on top of a cantilever. The radii of PC holes inside the ring are optimized in order to sense smaller forces in the range of 0-1µN with increments of 0.1 µN.  It is observed that a minimum force of 16 nN is required to cause a 0.1nm change in wavelength. Very high quality factor of 6500 is obtained in the studied force range. We  also report  the sensor characteristics in the force range of 0 to 0.1µN, 0.5 to 0.6µN and 0.9 to 1µN with force increments of 0.01µN and the minimum force needed for a 0.09nm change in wavelength is found to be 10nN. Fabrication details are also included.

 

Title Speaker Time & Venue Video (optional)
Information geometric robust estimation Prof. Rajesh Sundaresan, Dept. ECE., IISc Time: 4:00 pm (tea/coffee at 3:45pm). Date: 06th Mar. 2015 (Friday). Venue: Golden Jubilee Hall, ECE Dept. Link

Abstract:

The talk will be on the geometry associated with a robust variant of the maximum likelihood estimation procedure. Suppose that we are given independent and identically distributed samples of an unknown distribution. The distribution comes from a parametric family. The maximum likelihood estimate is the member of the family that is closest, in the sense of relative entropy or Kullback-Leibler divergence, to the empirical measure of the observations. One robust variant of the maximum likelihood estimate is the member of the family that is closest, not in the sense of relative entropy, but in the sense of a parametric generalisation of relative entropy. One may view the resulting distribution as a "projection" of the empirical distribution on the family. The talk will highlight the geometry associated with this projection. The talk will also discuss a simplified computation procedure, one that is suggested by the geometric view point, when the projection is on a power-law family. The talk is based on the Ph.D. work of M. Ashok Kumar.

 

Title Speaker Time & Venue Video (optional)
Magnetic nanoswimmers Prof. Ambarish Ghosh, Dept. ECE., IISc
Time: 4:00 pm (tea/coffee at 3:45pm). Date: 20th Feb. 2015 (Friday). Venue: Golden Jubilee Hall, ECE Dept. Link

Abstract:

 I will share some recent results from our group in the study of magnetically powered nanoswimmers, in particular highlight: (i) a new technique to maneuver them independently and (ii) a new technology to measure the rheological properties of a complex heterogeneous material. I will end the talk describing how the method used to fabricate the swimmers can be adapted to develop a new class of ultra-sensitive plasmonic photodetectors.

 

Title Speaker Time & Venue Video (optional)
An Explicit Sparse Recovery Scheme in the L1 norm Prof. Arnab Bhattacharyya, CSA Dept., IISc
Time: 4:00pm (tea/coffee at 3:45pm). Date: 14 Nov. 2014 (Friday). Venue: Golden Jubilee Hall, ECE Dept. Link

Abstract:

Consider the approximate sparse recovery problem: given Ax,where A is a known m-by-n dimensional matrix and x is an unknown (approximately) sparse n-dimensional vector, recover an approximation tox. The goal is to design the matrix A such that m is small and recovery is efficient. Moreover, it is often desirable for A to have other nice properties, such as explicitness, sparsity, and discreteness. In this work, we show that we can use spectral expander graphs to explicitly design binary matrices A for which the column sparsity is optimal and for which there is an efficient recovery algorithm (\ell_1-minimization). In order to recover x that is close to \delta, n-sparse (where \delta is a constant), we design an explicit binary matrix A that has m = O(sqrt{\delta} log(1/\delta)n) rows and has O(log(1/\delta)) ones in each column. Previous such constructions were based on unbalanced bipartite graphs with high vertex expansion, for which we currently do not have explicit constructions. In particular, ours is the first explicit non-trivial construction of a measurement matrix A such that Ax can be computed in O(n log(1/\delta)) time. Joint work with Vineet Nair.

 

Title Speaker Time & Venue Video (optional)
Codes for Distributed Storage - Asking More of an Old Friend Prof. P. Vijay Kumar, ECE Dept., IISc

Time: 4:00 pm (tea/coffee at 3:45 pm). Date: 31 Oct. 2014 (Friday). Venue: EC 1.08, ECE Dept.

Link
Abstract:
Given the explosive growth in the amount of data generated and the failure-prone nature of a storage device, there is renewed interest within the storage industry, in identifying techniques that will enable data to be stored in a manner that is efficient, reliable and which provides easy access to the data. Codes for distributed storage place data across nodes in a network in a redundant fashion and are designed precisely with these objectives in mind. This talk will focus on two recent approaches to code design that attempt to overcome the shortcomings of prevailing coding schemes that are based either on replication or else the use of Reed-Solomon codes. Quite apart from the application to distributed storage, these recent developments also represent fundamental advances in the theory of error-correcting codes. The talk will be tutorial in nature. (This talk was delivered earlier as a plenary talk at the 2014 IEEE International Symposium on Information Theory, July 4, Hawaii).

 

Title Speaker Time & Venue Video (optional)
Network Coding, Network-Error Correction and Matroids Prof. B. Sundar Rajan, ECE Dept., IISc Time: 4:00 pm (tea/coffee at 3:45 pm). Date: 17 Oct. 2014 (Friday). Venue: Golden Jubilee Hall, ECE Dept. Link
Abstract:
Network coding is a technique to increase the rate of information transmission through a network by coding (instead of simple routing) different information flows in the network. One of the main problems in network coding is to find whether a given network with a set of sources and sink demands is solvable using a scalar linear network code. Matroids are discrete objects which abstract the notion of linear dependence among vectors. They arise naturally in several discrete structures including graphs and matrices. The relationship between network coding and matroid theory (introduced by Dougherty, Freiling, and Zeger) shows that the scalar linear solvability of a network with a given set of demands is related to the existence of a representable matroid satisfying certain properties. Using the equivalence between networks and matroids several interesting results have been reported. Linear network-error correcting codes are special kinds of linear network codes which could correct errors that occur in the edges of the network. The bounds and constructions similar to classical error control coding techniques have been carried over to the context of linear network-error correction and detection. In this talk we present the connection between matroids and network-error correcting codes. Using this connection several construction and algorithmic results are presented for network-error correcting codes. [Joint work with Dr. Krishnan Prasad (IIIT Hyderabad). A full length version of this talk is to appear in IEEE Transactions on Information Theory- currently it is avialable as Arxiv:1201.6459v2 [cs.IT].

 

Title Speaker Time & Venue Video (optional)
Design of Sparse Antenna Arrays Prof. Sanjit Mitra, Univ. of California, Santa Barbara, and Univ. of Southern California

Time: 11:00 am (tea/coffee at 10:45 am). Date: 17 Oct. 2014 (Friday). Venue: Golden Jubilee Hall, ECE Dept.

Link
Abstract:
In this talk we present a general method for the factorization of a polynomial with unity coefficients into a product of sparse polynomials with fewer non-zero coefficients with unity values. The factorization method is then used to design sparse antenna arrays with uniform and linearly tapered effective aperture functions.

To see list of past colloquium: Click here.