Previous Talks2017-2018Annual Event Scheduled on April 18th 3-5PM Please vote for your favourite 3MT of 2017-2018 here.
Title: Finite Blocklength Analysis of Channel Capacity Date: March 14th, 2018 Abstract: Channel capacity is well understood as an asymptotic concept. That is, it is a limit that we can approach with large blocklengths. However, if the blocklength were to be restricted, then the maximum achievable rate is also affected and will be a function of the block-length. As is the case with channel capacity, you have an achievability and a converse analysis but they needn’t agree. Moreover, if you are good with bounds, you can show a gap of O(log n) for several channels. In this talk, we will study some tools used to compute finite blocklength lower as well as upper bounds on capacity. We will also see an application to a well known channel if time permits. This seminar is based on my research work in characterizing finite blocklength bounds for energy harvesting AWGN channels. Speaker Bio: Konchady Gautam Shenoy obtained his B.E in Electronics and Communications in NMAM Institute of Technology, Nitte in 2008. He then earned an M.Tech (with thesis) in Electrical Engineering dept. at IIT Bombay in 2011. He is currently pursuing a Ph.D. in ECE Dept. IISc under the guidance of Prof. Vinod Sharma. Gautam’s interests are in classical information theory and its varied applications, probability theory, mathematical analysis and puzzles (recreational math). Title: Representation of language in our brain Date: Febraury 14th, 2018 Abstract: Language is one of the key elements towards the progress of human beings. It helps us to transmit the information from one generation to another. Compared to vocalization, written form of language is a very recent cultural invention, and our brains could not have evolutionary adapted to process it since birth. The written form is processed in a part of the brain, popularly known as Visual Word Form Area (VWFA) and this are is formed only when we learn to read. In this talk, I will give an overall view of language network in the brain and more specifically talk about written form of language. I will try to cover (although not comprehensively) a vast literature on how learning to read leads to the formation of VWFA, its connectivity with the language network and models developed to understand the underlying representation of letters/ words. Speaker Bio: Aakash Agrawal completed his B.Tech in Electrical Engineering from NIT Hamirpur in 2014. After which he joined for a PhD program at Center for BioSystems Science and Engineering. He is jointly guided by S.P.Arun (CNS) and K.V.S. Hari (ECE). His research interests underlies in understanding how vision is processed in the brain. Using Behaviour and brain imaging (fMRI) techniques, he studies how reading changes letter representation in the brain. Title: Low-Delay Decentralised MAC protocols for Time-Slotted Collocated Wireless Nodes Date: Abstract: We consider a system of several collocated nodes sharing a time slotted wireless channel, and seek a MAC that (i) provides low mean delay, (ii) has distributed control (i.e., there is no central scheduler), and (iii) does not require explicit exchange of state information or control signals. The design of such MAC protocols must keep in mind the need for contention access at light traffic, and scheduled access in heavy traffic, leading to the long-standing interest in hybrid, adaptive MACs. We first propose EZMAC, a simple extension of an existing decentralized, hybrid MAC called ZMAC. Next we develop another protocol, QZMAC, using motivations from our extensions of certain delayoptimality and throughput-optimality theory from the literature. A method to improve the short-term fairness of QZMAC is proposed and analysed, and the resulting modified algorithm is shown to possess better fairness properties than QZMAC. The theory developed to reduce delay is also shown to work in the presence of transmission errors and fast fading. Extensions to handle time critical traffic (alarms, for example) and hidden nodes are also discussed. Practical implementation issues are outlined. Using simulations, we show that both protocols achieve mean delays much lower than those achieved by ZMAC, and QZMAC provides mean delays very close to the minimum achievable in this setting, i.e., that of the centralized complete knowledge scheduler. Speaker Bio: Avinash Mohan obtained his B.Tech. and M.Tech. degrees from the P.E.S Institute of Technology and the Indian Institute of Technology (IIT) Madras, respectively. He is currently a doctoral candidate in the Network Engineering Laboratory at the Indian Institute of Science, Bangalore. His research interests include resource allocation in wireless communication networks, stochastic control and reinforcement learning for wireless network performance optimization. Title: Quantum Computing and Quantum Communication Date: Abstract: Abstract_pdf Speaker Bio: Rohit Ramakrishnan is a PhD student working with Prof. T Srinivas. He received his M.Tech from International School of Photonics, CUSAT and B.Tech in Electronics and Communication Engineering from College of Engineering Trivandrum. He worked as researcher in Centre for Quantum Technologies, National University of Singapore and Australian Defence Force Academy before joining Indian Institute of Science, in 2014. His research focuses on High Dimensional Quantum Communication oriented towards its practical implementation. Title: Interference Alignment in Index Coding .Theory, Techniques and Results. Date: Abstract: Interference management through interference alignment is a well researched topic in communications. The relatively new idea of using interference alignment to index coding scenarios has inspired an interesting line of research. In an index coding scenario, the primary aim is to minimize the transmission bandwidth. When interference alignment is employed, the minimum number of independent dimensions required to avoid interference equates to the minimum number of transmissions of an optimal index code (minrank). The talk details on this concept and elaborates a new algorithm for finding the minrank of any symmetric index coding problem. Speaker Bio: Niranjana Ambadi is a PhD student working with Prof. B. Sundar Rajan. She received her B.Tech. in Electronics and Communication Engineering from National Institute of Technology Calicut in 2013. She worked as an Application Developer with Oracle India Private Limited before joining Indian Institute of Science, in 2014. Her areas of research include Network Coding, Index Coding and Matroid Theory. Title: Covariance Matching Techniques for Sparsity Pattern Recovery from Compressive Measurements. Date:September 13, 2017 Abstract: We consider the compressive sensing problem wherein multiple measurements vectors (MMVs) are used to find the common nonzero support of a joint sparse signal ensemble. Finding the nonzero support of signals is an important canonical problem encountered in multi-sensor signal processing applications such as cooperative spectrum sensing, direction of arrival estimation, target localization, distributed source coding etc. In this talk, we present the covariance matching framework which is capable of recovering O(m^2) sized supports using only m measurements per signal. This is in contrast to O(m) sized support recovery guarantees offered by conventional support recovery methods. The framework is linked to several existing support recovery algorithms. We also highlight its potential in spawning a new category of cost functions for sparse support recovery. The performance of the covariance matching based support recovery approach is linked to the Restricted Isometry Property (RIP) of the self-Khatri-Rao product of the measurement matrix used for generating the compressive measurements. We will present new results characterizing the RIP of a generic Khatri-Rao product matrix. Speaker Bio: Saurabh Khanna is a PhD student in the Dept. of ECE working with Prof. Chandra R. Murthy. His research interests are in the areas of structured signal processing, inverse problems and statistical learning theory. Currently, he is working on design and analysis of Bayesian techniques for sparse signal recovery. Title: Throughput-optimal discrete rate adaptation for threshold-based feedback in OFDM systems. Date: Slides Speaker Bio: Vineeth Kumar is a PhD student in ECE Dept, advised by Prof. Neelesh B. Mehta at the Indian Institute of Science. His current research is on rate adaptation and scheduling in OFDM systems with limited feedback. Title : On Kernelized Multi-armed Bandits Date: August 2, 2017 (Wednesday) Abstract : We consider the stochastic bandit problem with a continuous set of arms, with the expected reward function over the arms assumed to be fixed but unknown. We provide two new Gaussian process-based algorithms for continuous bandit optimization âmproved GP-UCB (IGP-UCB) and GP-Thomson sampling (GP-TS), and derive correspondd ing regret bounds. Specifically, the bounds hold when the expected reward function belongs to the reproducing kernel Hilbert space (RKHS) that naturally corresponds to a Gaussian process kernel used as input by the algorithms. Along the way, we derive a new self-normalized concentration inequality for vector-valued martingales of arbitrary, possibly infinite, dimension. Finally, experimental evaluation and comparisons to existing algorithms on synthetic and real-world environments are carried out that highlight the favorable gains of the proposed strategies in many cases. Speaker Bio : Sayak Ray Chowdhury is a PhD student in department of ECE working with Prof. Aditya Gopalan. Previously he did his ME in System Science and Automation from IISc. His research interest broadly lies in Machine Learning and Reinforcement Learning. Specifically, he is interested in Sequential Decision Making and Multi-armed Bandit problems with applications in sensor networks, recommendation systems and social networks. Title: Scene Text Recognition for Augmented Reality Date: 12 July 2017 (Wednesday) Abstract: Natural scene text recognition is an important aspect of scene understanding and could be a useful tool in building engaging augmented reality applications. In this work, we address the problem of false positives in text spotting. We propose improving the performace of sliding window text spotters by looking for character pairs (bigrams) rather than single characters. An efficient convolutional neural network is designed and trained to recognize bigrams. The proposed detector reduces false positive rate by 29.5% on the ICDAR 2015 dataset. We demonstrate that detecting bigrams is a computationally inexpensive way to improve sliding window text spotters. Speaker Bio: Sagar is a PhD student advised by Prof. Bharadwaj Amrutur at the Indian Institute Of Science. His current research is on natural scene text recognition. In a previous life, he worked on low-power circuits. Title: Distributed Control and Quality of Service in Multihop Networks Date: 14 June 2017(Wednesday) Abstract: A wireless network consists of a number of nodes, connected to each other by a time varying wireless medium. Various applications give rise to flows, which are routed through the nodes in the network. These flows will also have different Quality-of-Service (QoS) requirements; for example, constraints on the maximum delay it is willing to incur, minimum bandwidth requirement, and so on. A network controller would be interested in devising a network control scheme, consisting of scheduling, routing and power control, while simultaneously trying to meet the QoS demands. This talk will discuss a distributed randomized algorithm to address this problem in the context of multihop networks. Speaker Bio: Ashok Krishnan is a PhD student working at Performance analysis lab, Dept of ECE, IISc under the guidance of Prof. Vinod Sharma. He is currently interested in performance analysis of wireless multihop networks and quality of service. Title: Efficient recovery from Multiple Erasures by Accessing Small Number of Disks for Distributed Data Storage Date: This work is a finalist at Jack Keil Wolf Student Paper Award at ISIT 2017 Abstract: Traditionally, replication has been the technique used to provide reliability in distributed data storage. Erasure codes then entered the scene and proposed to reduce the storage overhead while maintaining reliability. If a small number of disks each storing a part of the data become inaccessible, it is necessary to restore the system to original state by reconstructing the contents of the failed disks. This being a system maintenance operation, it is desirable to be as efficient as possible in terms of usage of resources, not disturbing the regular operation of the storage system. Locality, is the name for one such desirable property where one repairs a set of failed disks by accessing as less disks as possible. An [n; k] code is said to be a locally recoverable code with sequential recovery from t erasures, if for any set of s <= t erasures, there is an s-step sequential recovery process, in which at each step, a single erased symbol is recovered by accessing at most r other code symbols. In this presentation, a tight upper bound on the rate of such a code, for any value of number of erasures t and any value r >= 3, of the locality parameter is presented. This bound proves an earlier conjecture due to Song, Cai and Yuen. A matching construction of binary codes that are rate-optimal is also presented. Recently majority-logic decodable codes re-appeared in the context of distributed storage, called by a different name “codes with availability”. We will briefly talk about deriving a rate and minimum distance bound for these codes. Bio: Balaji S.B is a PhD student and Ganesh Kini a Masters Student, both with Prof Vijay Kumar at CSD Lab, ECE, IISc. Their present research interests include codes with locality and availability. Title: The pattern maximum likelihood estimation problem. Date: 19th April 2017(Wednesday) Abstract: Property estimation of discrete distributions is an important problem that has received a lot of attention over the years. These problems are particularly interesting in the large-alphabet regime and several techniques have been proposed to estimate various properties such as entropy, support size, and distance to uniformity, with different techniques used to analyze each property. I will talk about the profile maximum likelihood (also called pattern maximum likelihood, or PML for short) estimation problem, introduced by Orlitsky et al. in the context of universal compression of sources over unknown alphabets. A recent result of Acharya et al. shows that a simple plug-in estimator derived from the PML estimate is competitive, and can be used to near-optimally estimate any symmetric property of the distribution. I will also indicate generalizations to Markov sources, the connection between the PML estimation problem and the permanent of a certain matrix, and algorithms to approximate these estimates efficiently. Bio: Shashank Vatedka received his BE from PES Institute of Technology in 2011 and his PhD from the department of ECE, IISc in April 2017. Since September 2016, he is a research assistant at the Institute of Network Coding, Chinese University of Hong Kong. His research interests are coding theory, information theoretic security, lattices, graphical models, and iterative decoding. |