SPCOM2018

Welcome to SPCOM 2018!

TECHNICAL CO-SPONSORS

IEEE IEEE Signal Processing Society

TUTORIAL SESSIONS

All tutorials will be held on Monday, July 16 with three parallel sessions each in the morning and the afternoon, at the Department of ECE.

TITLE TUTOR TIME PLACE SLIDES
Teasing out the multi-scale representational space of cross-modal speech perception: Methods and mechanisms Arpan Banerjee (NBRC) 09:30-13:00 ECE 1.07

slides1

slides2

5G: An Evolution Towards a Revolution Karthik Sundaresan (NEC Labs) 09:30-13:00 ECE 1.08 slides
PCA and Robust PCA for Modern Datasets Namrata Vaswani (Iowa State) 09:30-13:00 ECE GJH slides
Need for Low Power Communications and Localization Dinesh Bharadia (UCSD) 14:15-17:45 ECE 1.07 slides
Video Streaming: On Rate-Adaptation, Multipath, Virtual Reality, and Content Distribution Network Vaneet Aggarwal (Purdue) 14:15-17:45 ECE 1.08 slides
Introduction to Reinforcement Learning Harm Van Seijen (Microsoft Research) 14:15-17:45 ECE GJH slides

Teasing out the multi-scale representational space of cross-modal speech perception: Methods and mechanisms

Arpan Banerjee

 

Speaker:  Arpan Banerjee

National Brain Research Center, India


Bio: Arpan Banerjee received his PhD in Complex Systems and Brain sciences from Florida Atlantic University, USA primarily working in the area of bimanual motor coordination in humans. He has completed his post-PhD training at Center for Neural Sciences, New York University and The National Institutes of Health, USA working in signal processing and spike train, LFP and MEG recordings. Currently his interests are in using computational neuroscience and multimodal brain imaging EEG/MEG/ fMRI to understand accurately where (spatial) and when (temporal) task-related differences in information processing occur in the brain during multisensory integration, higher order visual processing and cognition. The key research question that he wants to address is how large networks of neurons coordinate amongst each other to form organized assemblies at only specific instants of time to orchestrate ongoing behavior. Demystifying the tunes that govern this neural orchestra will shed light to subtle differences in human brain function across normal individuals, across patients and eventually lead to developing neuro-markers for spectrum disorders such as autism.

 Abstract: Multisensory integration has excited a large group of researchers from psychologists, computer scientists, neurphysiologists and finally neuroimaging community and triggered a wide body of research. Yet, the representational space of multisensory processing such as during cross-modal speech perception remains elusive. In this talk I would like to delimit the boundaries of this representational space using the results obtained from multimodal neuroimaging techniques, EEG and fMRI.

In the first part of this talk I will talk about network analysis tools that are currently used in the literature for analysis of EEG/ MEG and functional MRI data. Network methods have become an important tool to identify and characterize neural mechanisms of various cognitive process as well as quantifying neurological and neuropsychiatric disorders. I will also present some existing issues with EEG/ MEG source analysis techniques and discuss the use of these methods with empirical data sets. I will talk about neurobiologically realistic modeling tools using dynamic systems theory. Thereafter I will illustrate how the latter approach is important in interpreting the outcome of network analysis tools in particular validation of ground truth.

In the second part of the talk, I would talk about a behavioral paradigm with which we have been able to track cross-modal speech perception using psychophysical control parameters such temporal ordering of audio-visual stimulus. Using EEG and fMRI recordings on human volunteers, I will illustrate how the spatiotemporal functional network patterns can be used to understand the processing of behavior. Finally I would present a computational model inspired by neurobiologically realistic parameters that attempts to link the behavioral results with patterns of activity observed in neuroimaging recordings. The overarching goal of the talk is to build a mechanistic understanding of the neural dynamics observed at individual brain regions and across a functional network comprised of multiple brain areas underlying speech perception.

PCA and Robust PCA for Modern Datasets

Namrata Vaswani

 

Speaker:  Namrata Vaswani

Iowa State University, USA


Bio: Namrata Vaswani is a Professor of Electrical and Computer Engineering, and (by courtesy) of Mathematics, at Iowa State University. She received a Ph.D. in 2004 from the University of Maryland, College Park and a B.Tech. from Indian Institute of Technology (IIT-Delhi) in India in 1999. Her research interests lie at the intersection of statistical machine learning / data science, computer vision, and signal processing. She is a recipient of the Harpole-Pentair Assistant Professorship and the Iowa State Early Career Engineering Faculty Research Award at Iowa State. In 2014, she received the IEEE Signal Processing Society (SPS) Best Paper Award for her Modified-CS work that was co-authored with her graduate student Lu in the IEEE Transactions on Signal Processing in 2010. Vaswani has served the SPS and IEEE in various capacities. She is an Area Editor for IEEE Signal Processing Magazine and has served twice as an Associate Editor for IEEE Transactions on Signal Processing. She is the Lead Guest Editor for a Proceedings IEEE Special Issue on Rethinking PCA for Modern Datasets, and of a Signal Processing Magazine Feature Cluster on Exploiting Structure in High-dimensional Data Recovery, both of which will appear in 2018.She is also the Chair of the Women in Signal Processing (WiSP) Committee, a steering committee member of SPS's Data Science Initiative, and an elected member of the SPTM and IVMSP Technical Committees.

 Abstract: In today's big data age, there is a lot of data generated everywhere around us. Examples include texts, tweets, network traffic, changing Facebook connections, or video surveillance feeds coming in from one or multiple cameras. Before processing any big dataset, the first step is to perform dimension reduction and noise/outlier removal. Traditionally, dimension reduction is done by solving the principal components' analysis (PCA) problem. While this is a very old problem, many of the traditional techniques fail if the data is corrupted by anything other than small and uncorrelated noise. PCA and robust PCA and their streaming counterparts have a very large number of applications since dimension reduction is a key first step in a very large variety of applications. Some examples include exploratory data analysis, video analytics, recommendation system design, and many more.

The tutorial will begin with a brief introduction to the basic random matrix theory results needed by some of the theoretical guarantees that will be discussed (depending on audience background and interest). Most of the tutorial will talk about the original PCA problem; about PCA when data and noise are correlated (correlated-PCA); and about PCA in the presence of large but structured, e.g., sparse, noise (robust PCA). Moreover, because all the data cannot be stored, or because there is a need to make decisions in real-time, and/or because the structure of the data could itself change significantly over time, there is a lot of interest in streaming algorithms for PCA or robust PCA and their dynamic (tracking) counterparts. About half of the tutorial will talk about old and new approaches to streaming PCA and streaming dynamic robust PCA.

PCA has been a problem that has been studied for almost a century dating back to the work of Hotelling from the 1930s among others. However, the correlated-PCA problem has received almost no attention until very recently. Robust PCA has also been studied for a few decades. However, the new series of works on provably correct and practically usable robust PCA started appearing in 2011 and later. The work on provably correct streaming or dynamic robust PCA techniques only started appearing in 2014 and later. There has been older work on streaming or online PCA, but there has been much a renewed interest in recent years on online PCA, streaming (memory-optimal single pass) solutions for PCA, and on fast algorithms for partial SVD.

5G: An Evolution Towards a Revolution

Karthik Sundaresan

 

Speaker:  Karthik Sundaresan

NEC Laboratories America, USA


Bio: Karthik Sundaresan is a senior researcher in the mobile communications and networking research department at NEC Labs America. His research interests are broadly in wireless networking and mobile computing, and span both algorithm design as well as system prototyping. He is the recipient of ACM Sigmobile’s Rockstar award (2016) for early career contributions to the field of mobile computing and wireless networking, as well as several best paper awards at prestigious ACM and IEEE conferences. He holds over thirty patents and received a business contribution award from NEC for the technology commercialization of an LTE small-cell interference management technology. He has participated in various organization roles for IEEE and ACM conferences, and served as the PC co-chair for ACM MobiCom’16. He is a senior member of IEEE and currently serves as an associate editor for IEEE Transactions on Mobile Computing.

 Abstract: The aim of this tutorial is to give the audience an overview of the landscape of the future generation of mobile networks, namely 5G. Contrary to popular view, 5G is not expected to be anchored on a single disruptive technology but rather supported by an amalgamation of multiple technologies. In essence, it is an evolution of several key technical advancements, whose synergy is expected to revolutionize the heterogeneity of use cases that can be “simultaneously” enabled by a single network. Such use cases range from throughput-focused mobile broadband (Gigabit peak rates) to latency/reliability-focused mission critical (e.g. augmented/virtual reality, autonomous driving) and density-focused massive connection (IoT) services.

While physical layer advancements in the form of New Radio (NR) are an integral part of 5G, realizing the diverse use cases envisioned, will equally require innovation and flexible orchestration of its access, network and computing layers as well. This tutorial will provide an overview of some of these innovative ingredients that will constitute 5G, from the perspective of not just radio access network, but also core network and services/applications. It will cover topics ranging from communication and networking to architectural designs, automation and use cases, including but not limited to

1. Radio: new radio (NR), mmWave, flexible OFDM numerology, advanced coding 

2. Access: IoT-optimized access, hybrid access (licensed, shared and unlicensed 
spectrum)
3. Network: cloud deployments, virtualization and network slicing 

4. Computing: network function virtualization, scalable core design, mobile edge 
computing
5. Automation: network access, provisioning and management
6. Case studies: Augmented reality over LTE networks, self-configuring UAV-based LTE 
networks 


Introduction to Reinforcement Learning

Harm van Seijen

 

Speaker:  Harm van Seijen

Microsoft Research, Montreal, Canada


Bio: Harm van Seijen is the research manager of the reinforcement learning team at Microsoft Research, Montreal. His work focuses on fundamental challenges in reinforcement learning. He obtained his PhD in 2011 from the University of Amsterdam on the topic of reinforcement learning under space and time constraints. Prior to his position at Microsoft, he was a Postdoctoral Fellow at the University of Alberta, working together with Professor Richard Sutton on novel reinforcement-learning methods, and was affiliated with the startup company Maluuba.

 Abstract: This talk will give an overview of reinforcement learning, a machine learning approach to learn optimal behavior that has gained a lot of traction in the last few years. In the reinforcement-learning setting, an agent interacts with an initially unknown environment and tries to maximize the total reward it receives via a trial-and-error process. By using deep neural networks as internal representation, reinforcement learning methods have become substantially more powerful in recent years, achieving above-human performance on many challenging tasks, from robotic control to the ancient game of Go.  We will discuss the basic theory behind reinforcement learning and discuss the relation with other popular machine learning approaches. Furthermore, we will discuss recent results, as well as remaining challenges and active areas of research.

Video Streaming: On Rate-Adaptation, Multipath, Virtual Reality, and Content Distribution Network

Vaneet Aggarwal

 

Speaker:  Vaneet Aggarwal

Purdue University, USA


Bio: Vaneet Aggarwal received the B.Tech. degree in 2005 from the Indian Institute of Technology, Kanpur, India, and the M.A. and Ph.D. degrees in 2007 and 2010, respectively from Princeton University, Princeton, NJ, USA, all in Electrical Engineering.

He is currently an Assistant Professor at Purdue University, West Lafayette, IN (2015-current) and a VAJRA Adjunct Professor at IISc Bangalore (2018-current). Prior to this, he was a Senior Member of Technical Staff Research at AT&T Labs-Research, NJ (2010-2014), and an Adjunct Assistant Professor at Columbia University, NY (2012-2014). He is an IEEE Senior Member (2015-current). His current research interests are in communications and networking, video streaming, cloud computing, and machine learning.

Dr. Aggarwal is on the editorial board of the IEEE Transactions on Communications and the IEEE Transactions on Green Communications and Networking. He was the recipient of Princeton University's Porter Ogden Jacobus Honorific Fellowship in 2009, the AT&T Key Contributor award in 2013, AT&T Vice President Excellence Award in 2012, and AT&T Senior Vice President Excellence Award in 2014. He was also the recipient of the 2017 Jack Neubauer Memorial Award, recognizing the Best Systems Paper published in the IEEE Transactions on Vehicular Technology.

 Abstract: Mobile video has emerged as a dominant contributor to cellular traffic. It already accounts for around 40-55 percent of all cellular traffic and is forecast to grow by around 55 percent annually through 2021. While its popularity is on the rise, delivering high quality streaming video over cellular networks remains extremely challenging. In particular, the video quality under challenging conditions such as mobility and poor wireless channel is sometimes unacceptably poor. Almost every viewer at some point in time can relate to experiences of choppy videos, stalls, etc. This tutorial aims to provide fundamental approaches to improve the quality of experience (QoE) for video viewing at the end users. 

Not surprisingly, a lot of attention from both research and industry in the past decade has focused on the development of adaptive streaming techniques for video on demand that can dynamically adjust the quality of the video being streamed to the changes in network conditions. In this tutorial, we will start with explaining the basics of adaptive bit-rate video streaming, and some of the existing algorithms. Further, we will theoretically formulate the problem of adaptive video streaming with the knowledge of future bandwidth. The non-convex integer-constrained streaming problem will be showed to be solvable optimally in linear time complexity, giving a new class of algorithms in combinatorial optimization which in complexity class P. The algorithms can be extended to window-based online mechanisms, with harmonic mean or crowd-sourced imperfect bandwidth prediction. Results over a realistic testbed will also be demonstrated. Further extensions to multiple paths, and link preference (eg. WiFi over LTE) will be provided.

The 360-degree technology is shaping the video industry. 360-degree videos provide users a panoramic view creating a unique viewing experience. 360-degree videos, also known as immersive or spherical videos, are essential parts of the virtual reality (VR) which are changing the user’s experience of video streaming. VR is projected to form a big market of $120 billion by 2020. 360-degree videos are very popular on major video platforms such as YouTube, Facebook. However, the current popular technologies for streaming try to fetch the all the portion of the chunk in the same quality including both the visible and invisible portions. Though this method is simple, it has some disadvantages. For example, the bandwidth utilization is high as the chunks in the 360-degree videos are of larger sizes compared to the traditional ones. Thus, if the network is congested or the bandwidth is low, it will lead to a poor-quality video. Hence, without smart algorithms, it can easily consume the wireless bandwidth. Even the wireline capacity may not be enough for such 360-degree videos. We will provide the challenges in designing bandwidth-efficient streaming algorithm for 360-degree videos for maximizing the quality of service (or quality of experience) of the users. Such approaches use head movement prediction, which brings new challenges in addition to the bandwidth prediction.

So far, we considered the aspect of the last hop, which is wireless. In the final part, we will present the network side of the video transfer. The network designers can only control the wired part. With the same network controlling multiple users, the network becomes a bottleneck. We will provide a holistic framework considering the multiple network control knobs to optimize delivery from the network. Over-the-top video streaming, e.g., Netflix and YouTube, has been dominating the global IP traffic in recent years. The traffic will continue to grow due to the introduction of even higher resolution video formats such as 4K on the horizon. As end-users consume video in massive amounts and in an increasing number of ways, service providers need flexible solutions in place to ensure that they can deliver content quickly and easily regardless of their customer’s device or location. More than 50% of over-the-top video traffic is now delivered through content distribution networks (CDNs). Even though multiple solutions have been proposed for improving congestion in the CDN system, managing the ever-increasing traffic requires a fundamental understanding of the system and the different design flexibilities (control knobs) to make the best use of the hardware limitations. The service providers typically use two-tiered caching approach to improve the streaming service. In addition to the distributed cache servers provided by the CDN, the edge router can also have a cache. The different control knobs include the choice of distributed server, caching, queue management, etc., to optimize the end user QoE.

Need for Low Power Communications and Localization

Dinesh Bharadia

 

Speaker:  Dinesh Bharadia

University of California San Diego, USA


Bio: Dinesh Bharadia is faculty in ECE at University of California San Diego. Prior to UCSD, Dinesh Bharadia received his Ph.D. from Stanford University was a Postdoctoral Associate at MIT. Specifically, in his dissertation, he built the prototype of a radio, that invalidated a long-held assumption in wireless is that radios cannot transmit and receive at the same time on the same frequency. In recognition of his work, Dinesh was named to Forbes 30 under 30 for the science category worldwide list. Dinesh was also named a Marconi Young Scholar for outstanding wireless research and awarded the Michael Dukakis Leadership award. He was also named as one of the top 35 Innovators under 35 in the world by MIT Technology Review in 2016. Dinesh is also the recipient of the Sarah and Thomas Kailath Stanford Graduate Fellowship. From 2013 to 2015, he was a Principal Scientist for Kumu Networks, where he worked to commercialize his research on full-duplex radios, building a product that underwent successful field trials at Tier 1 network providers worldwide like Deutsche Telekom and SK Telecom. This product is currently under deployment. His research interests include advancing the theory and design of modern wireless communication systems, wireless imaging, sensor networks and data-center networks.

 Abstract: Low Power communication and localization has applications in sensing and measuring of our environment, to building smart cities and smart transportation systems and so on. In this tutorial, I would present a communication and localization system which can connect to existing WiFi infrastructure while using low power backscatter techniques. Specifically, I would elaborate use of network coding principles to build the above communication system. I would show a real-time demonstration of the low power system using an embedded systems platform built during the project.