Welcome to the Signal and Information Processing Seminar Series at Rutgers!

The SIP Seminar Series at Rutgers University–New Brunswick brings together a diverse group of researchers, both from within and outside Rutgers, on (approximately) a biweekly basis to discuss recent advances in signal and information processing. The term of “Signal and Information Processing” used within the SIP Seminars is rather broad in nature and subsumes signal processing, image processing, statistical inference, machine learning, computer vision, control theory, harmonic analysis, information theory, etc.

Seminar Mailing List: You can subscribe to the SIP Seminars mailing list by sending an email to ECE_SIP-request@email.rutgers.edu with SUBSCRIBE in the subject of the email.

Fall 2021 Seminar Schedule: The SIP Seminars in Fall 2021 will be hosted by Prof. Emina Soljanin and will take place virtually on the following dates (Tuesdays) at 2 pm on Zoom: September 21; October 5 and 19; and November 2, 16, and 30.

Fall 2021 SIP Seminars (Host: Prof. Soljanin)

Dr. Swanand Kadhe

Title:FastSecAgg: Scalable Secure Aggregation for Privacy-Preserving Federated Learning

Abstract: In modern large-scale machine learning, federated learning has emerged as an important paradigm, where the training data remains distributed over a large number of clients (e.g., mobile phones, smart devices). In federated learning, each client trains a neural network model locally using their data, and the central server aggregates these local models to obtain an improved model. However, recent attacks have demonstrated that model parameters shared by clients can leak significant amounts of information about their training data, making privacy preservation a critical concern.

In this talk, I will present a secure aggregation protocol, FastSecAgg, that enables the central server to average local models in a privacy-preserving manner while being robust to client dropouts. FastSecAgg reduces the computation cost at the server by several orders-of-magnitude compared to the state-of-the-art schemes, and guarantees security against the server colluding with any subset of some constant fraction (e.g. ~10 %) of clients in the honest-but-curious setting. I will highlight the main building block of FastSecAgg — a novel multi-secret sharing scheme, FastShare, powered by the (finite field) Fast Fourier Transform (FFT). Finally, I will demonstrate that FastSecAgg achieves similar accuracy as vanilla federated averaging on LEAF benchmark datasets for federated learning.

Biography: Swanand Kadhe is a postdoctoral researcher in the EECS Department at the University of California Berkeley. He earned his Ph.D. degree in Electrical and Computer Engineering from Texas A&M University in 2017. He is a recipient of the 2016 Graduate Teaching Fellowship from the College of Engineering at Texas A&M University. He has been a visiting researcher at Nokia Bell Labs, Duke University, and The Chinese University of Hong Kong. From 2009 to 2012, he was an R&D engineer at the TCS Innovation Labs, Bangalore. His research interests lie broadly in federated and distributed machine learning, information and coding theory, privacy and security, and blockchains.

Prof. Hessam Mahdavifar

Title: Machine Learning-Aided Channel Coding: Opportunities and Challenges

Abstract: Today, channel codes are among the fundamental parts of any communication system, including cellular, WiFi, and deep space, among others, enabling reliable communications in the presence of noise. Decades of research have led to breakthrough inventions of various families of channel codes. Yet no unified approach exists in answering these two fundamental questions: Given a channel, how do we efficiently construct the best possible code? And given a channel code, how do we design an efficient and optimal decoder? In this talk, we will discuss how the remarkable advancements in data-driven machine learning (ML) can be leveraged toward answering these questions. In particular, we will focus on a class of codes rooting in Plotkin recursive construction. This class includes Reed–Muller (RM) codes as the state-of-the art binary algebraic codes, as well as polar codes, the first capacity-achieving codes with explicit, i.e., non-randomized, constructions. In the first part of this talk, we will present an efficient and close-to-optimal decoder obtained for RM codes by learning a pruning process applied to an exponentially complex decoder. In the second part, we will tackle the fundamental problem of designing new channel codes. In particular, we will demonstrate KO codes, a new class of channel codes designed by training neural networks while preserving Plotkin-like structures. KO codes beat both of their RM and polar code counterparts, under the successive cancellation decoding, in the challenging short-to-medium blocklength regime. We will also discuss various challenges that should be overcome to pave the way for adopting such ML-aided channel coding strategies in practice.

Biography: Hessam Mahdavifar is an Assistant Professor in the Department of Electrical Engineering and Computer Science at the University of Michigan. He received the B.Sc. degree from Sharif University of Technology in 2007, and the M.Sc. and the Ph.D. degrees from the University of California San Diego (UCSD) in 2009 and 2012, respectively, all in Electrical Engineering. He was with the Samsung Mobile Solutions Lab between 2012 and 2016. His general research interests are in coding and information theory with applications to wireless communications, machine learning, and security. He has won several awards including the NSF CAREER award in 2020, the Best Paper Award in the 2015 IEEE International Conference on RFID, the UCSD Shannon Memorial Fellowship, and two Silver Medals at the International Mathematical Olympiad.

Prof. Lara Dolecek

Title: Overcoming Data Availability Attacks in Blockchain Systems: A Graph-Coding Perspective

Abstract: Blockchain systems are already gaining popularity in a variety of applications due to their decentralized design that is favorable in many settings. To overcome excessive storage and latency burden, light nodes and side blockchains have been proposed to, respectively, enhance the basic blockchain architecture. However, both light nodes and side chains are vulnerable to data availability (DA) attacks by malicious nodes.  Recently, a technique based on erasure codes called Coded Merkle Tree (CMT) was proposed by Yu et al. that enables light nodes to detect a DA attack with high probability. CMT method relies on the use of random LDPC codes. We build on the previous work and demonstrate that graph codes specifically designed for the target applications in blockchain systems perform better than randomly constructed codes; intriguingly, the new finite-length code optimization framework unveils code properties beyond the established metrics.

Biography: Lara Dolecek is a Full Professor with the Electrical and Computer Engineering Department and Mathematics Department (courtesy) at the University of California, Los Angeles (UCLA). She holds a B.S. (with honors), M.S. and Ph.D. degrees in Electrical Engineering and Computer Sciences, as well as an M.A. degree in Statistics, all from the University of California, Berkeley. She received several awards for her research and teaching including the David J. Sakrison Memorial Prize from UC Berkeley, NSF CAREER Award, IBM Faculty Award, Okawa Research Grant and the Northrop Grumman Excellence in Teaching Award from UCLA. With her research group and collaborators, she received numerous best paper awards. She currently serves as an Associate Editor for IEEE Transactions on Information Theory and as the Secretary of the IEEE Information Theory Society. Prof. Dolecek is a 2021-2022 Distinguished Lecturer of the IEEE Information Theory Society. Prof. Dolecek has served as a consultant for a number of companies specializing in data communications and storage. In her current research, she is especially excited to explore the role of channel coding methods in blockchain systems, quantum information systems, and distributed storage and computing.

Dr. Alexei Ashikhmin

Title: TBD

Abstract: TBD

Biography: TBD

Prof. Rashmi K. Vinayak

Title: TBD

Abstract: TBD

Biography: TBD

Prof. Hossein Pishro-Nik

Title: TBD

Abstract: TBD

Biography: TBD