HOME

Attending

Call for Papers

Awards

Committees

Tutorials

Workshops

Invited Speakers

Plenary Speakers

Program

Paper Submission

Registration

Sponsors

Contact Us

About IIT Bombay


Financial Sponsors

Technical Co-Sponsors

Tutorials are complimentary this year and can be attended virtually without registering for the conference (details below).


1.Variations on a theme by Reed-Solomon: Applications to Distributed Storage

Speaker: Lalitha Vadlamani (IIIT Hyderabad)

Date/Time: May 23, 9:30 am - 12:00 pm

Link:https://zoom.us/j/95705779256?pwd=aUlWOHQ1Sld0TW80Vkt5Y3MrQ003QT09

Meeting ID: 957 0577 9256

Passcode: 210917

Abstract: Conventionally, distributed storage systems employed maximum distance separable codes(MDS) to recover data in the event of node failures. The most widely used class of MDS codes in practice are Reed-Solomon codes. Though MDS codes are efficient in terms of storage overhead, they cannot be efficiently repaired in case of single node failures using naive repair schemes. This tutorial focuses on different classes of codes (listed below) which allow for efficient repair of single node failures.

(i) Regenerating codes are a class of codes which minimize the repair bandwidth incurred in repairing a single failed node. A product-matrix framework will be introduced and optimal constructions of regenerating codes based on this framework will be discussed.

(ii) Reed-Solomon codes themselves can be efficiently repaired if the code symbols of the code are considered as vectors over a subfield. In this tutorial, trace-repair framework will be introduced which allows for efficient repair of Reed-Solomon codes. Also, an optimal construction of Reed-Solomon codes which achieve the cut-set bound will be presented.

(iii) Locally repairable codes (LRC) minimize the number of nodes contacted during repair. An optimal LRC construction popularly known as Tamo-Barg code will be discussed in detail. Tamo-Barg codes can be thought of as special subcodes of Reed-Solomon codes.

(iv) Maximally recoverable codes (MRC) are a class of codes which can correct the maximum possible number of erasure patterns, given the locality constraints of the code and hence of interest. Earlier constructions of maximally recoverable codes were based on linearized polynomials and Gabidulin codes. More recent constructions of MRC are based on linearized Reed-Solomon codes which are maximum sum-rank distance codes. Both these constructions will be presented.

2.An introduction to quantum information theory

Speaker: Pranab Sen (TIFR)

Date/Time: May 23, 9:30 am - 12:00 pm

Link:https://zoom.us/j/93455880980?pwd=UEFPSHluRjNKcXJJb3hpblkvOHl1QT09

Meeting ID: 934 5588 0980

Passcode: 959336

Abstract: Information carrying properties of quantum systems were first investigated in a seminal paper of Holevo in 1973, though earlier works by Belavkin dating back to 1967 had some information theoretic aspects too. However, it took the emergence of famous theoretical papers on quantum algorithms like Shor's integer factoring in 1994 for the rebirth of general interest in information theory of quantum systems. Works by Schumacher, Holevo, Westmoreland, Shor, Lloyd, Devetak and others in the late 1990s quickly managed to obtain fundamental results like the definition of quantum bits (qubits), entanglement bits (ebits), capacity of transmitting classical as well as quantum messages over point to point quantum channels etc., putting the infant quantum information theory on a firm footing.

In this tutorial, we will start with a rapid introduction to the basics of quantum information processing largely following the above historical development. Along the way we will briefly see some unique quantum features like teleportation, superdense coding, secure key distribution and information locking. We will then dive into a deeper study of some aspects of quantum Shannon theory, pointing out key differences with the classical theory. We will end with a fleeting view of recent advances.

3.Introduction to 5GNR standard

Speaker: Radhakrishna Ganti (IIT Madras)

Date/Time: May 24, 9:30 am - 12:00 pm

Link:https://zoom.us/j/99296457383?pwd=ZEJTS2pvdUFjeWZHNlh0WG53K245Zz09

Meeting ID: 992 9645 7383

Passcode: 373069

Abstract: In this tutorial, we will go through all the important aspects of 5GNR with an emphasis on the standard. In particular, we will cover all the downlink and uplink channels (Data, control) and the important physical layer procedures. The tutorial starts with the basics of time-frequency structure of 5G NR and builds on it. We will look at the following topics in detail

1) Time-Frequency structure

2) PBCH/PRACH sync channels

2) PDSCH/PUSCH (data channel) and DMRS pilots.

3) MIMO for PDSCH

4) PDCCH/PUCCH (Control channels)

4. Conformal prediction: a wrapper for quantifying predictive uncertainty of black box ML algorithms

Speaker: Aaditya Ramdas (Carnegie Melon University)

Date/Time:May 24, 12:30 pm - 3:00 pm

Link: https://zoom.us/j/99296457383?pwd=ZEJTS2pvdUFjeWZHNlh0WG53K245Zz09

Meeting ID: 992 9645 7383

Passcode: 373069

Abstract: Conformal prediction is a simple, popular, modern technique for providing valid predictive inference for arbitrary machine learning models. Its validity relies on the assumptions of exchangeability of the data, and symmetry of the given model fitting algorithm as a function of the data. I will spend the majority of the tutorial describing this framework, focusing on recent advances.

However, exchangeability is often violated when predictive models are deployed in practice. For example, if the data distribution drifts over time, then the data points are no longer exchangeable; moreover, in such settings, we might want to use an algorithm that treats recent observations as more relevant, which would violate the assumption that data points are treated symmetrically. Time permitting, I will describe a generalization of conformal methods to deal with both aspects. These algorithms are provably robust, with substantially less loss of coverage when exchangeability is violated due to distribution drift or other challenging features of real data, while also achieving the same coverage guarantees as existing conformal prediction methods if the data points are in fact exchangeable.

All along, we demonstrate the practical utility of these ideas with simulations and real-data experiments.

5. Adaptive Online Quickest Change Detection

Speaker: Aditya Gopalan (IISc)

Date/Time:May 24, 3:15 pm - 6:00 pm

Link: https://zoom.us/j/99296457383?pwd=ZEJTS2pvdUFjeWZHNlh0WG53K245Zz09

Meeting ID: 992 9645 7383

Passcode: 373069

Abstract: Industrial and surveillance applications employ a range of sensors that can gather vast amounts of data to monitor systems of interest for anomalous behaviour. In a streaming setting, practical constraints on compute and/or sensing often prohibit processing all the data sampled from all sensors at all times, and instead necessitate adaptive sampling for inference. Imagine, for example, monitoring hundreds of CCTV feeds to detect suspicious activity arising at any one location -- with access to a sophisticated, but slow, computer vision pipeline that can examine only a few feeds, which feeds would you attend to over time in order to narrow down on the anomalous location? We will introduce the formulation of bandit (i.e., adaptive & sequential) quickest change detection (BQCD), give an overview of traditional (non-adaptive) quickest change detection ideas, survey some recent algorithmic proposals for BQCD and theoretical performance analyses, and present numerical results to illustrate the performance gains from carefully designed sampling schemes. Time permitting, we will also describe connections to problems in nonstationary multi-armed bandits and open questions in this line of work.

This is partially based on joint work with Venkatesh Saligrama (Boston Univ.) and Braghadeesh Lakshminarayanan (ex-IISc).

6. The concentration of information phenomenon and its applications

Speaker:Mokshay Madiman (University of Delaware)

Date/Time:May 24, 6:30 pm - 9:30 pm

Link:https://zoom.us/j/99296457383?pwd=ZEJTS2pvdUFjeWZHNlh0WG53K245Zz09

Meeting ID: 992 9645 7383

Passcode: 373069

Abstract: The classical Shannon-McMillan-Breiman theorem expresses the fact that data from a stationary ergodic source lives with high probability in a “typical set” that is much smaller than the actual support of the process. In the last decade, this phenomenon has been found to be exhibited in a variety of contexts outside of the classical one. The story began with the discovery by Bobkov and the speaker that the information content per coordinate of data from a log-concave measure is highly concentrated. That result has since been sharpened and generalized in a number of directions, and some surprising applications have been developed. We will attempt to sketch this story in a cohesive manner, giving enough details for solid understanding and trying to present the cleanest and most illuminating proofs.


Financial Sponsors Technical Sponsors