September 8–11, 2020
10 am to 2 pm PT (1 pm to 5 pm ET) Daily
Watch Workshop Videos: C3.ai DTI YouTube Channel (From our homepage, scroll down to Workshops)
Epidemics spread through networks of human contact. Such networks have been studied for close to a century, and understanding them has never been more important as we face the COVID-19 pandemic.
Opinions are formed similarly through social interactions, in person or through the print and broadcast media, and increasingly over social media, where “information dynamics” can quickly influence perceptions and shape behavior. It has become increasingly urgent to understand how the spread of information through social media platforms, driven by bots, bad actors, and unsuspecting users, are creating networks that can result in extreme polarization, information echo chambers, and the proliferation of “fake news.”
Distributed engineered systems achieve desired performance through sharing information over component-connected communication networks. These networks have enabled the development of distributed databases, distributed computational frameworks, and more generally distributed autonomous agents.
These three types of networks, though different physically, share “dynamical” similarities at their core. At the very least, scientific studies about them have demonstrated similar models and analytic tools. This C3.ai DTI Workshop will bring together experts from a range of disciplines to help build common foundations about such networks as well as decipher their differences.
ORGANIZERS
Devavrat Shah (Massachusetts Institute of Technology), Lei Ying (University of Michigan)
SPEAKERS
Noah Friedkin (University of California, Santa Barbara), Matthew O. Jackson (Stanford University), Jon Kleinberg (Cornell University), Naomi Ehrich Leonard (Princeton University), Laurent Massoulie (Inria, Microsoft Research), Elchanan Mossel (Massachusetts Institute of Technology), H. Vincent Poor (Princeton University), Weina Wang (Carnegie Mellon University), Tauhid Zaman (Yale University)
PROGRAM
(All times are Pacific Time)
Day 1: Tutorials
ABSTRACT
The proliferation of fake news on online social networks has eroded the public trust in news media and has become an imminent threat to the ecosystem of online social platforms. Epidemics like the COVID-19 pandemic have devastating impacts on the economy and society. This tutorial will cover several inverse problems related to combating epidemics and online fake news using random graphs and machine learning, including how to locate the source of epidemics or fake news with partial observations, and how to reconstruct the history of epidemic/fake news spreading?
SPEAKER
Lei Ying is a Professor in the Electrical Engineering and Computer Science Department at the University of Michigan, Ann Arbor. Lei’s research is broadly in the interplay of complex stochastic systems and big-data, including large-scale communication/computing systems for big-data processing, private data marketplaces, and large-scale graph mining. He coauthored books Communication Networks: An Optimization, Control and Stochastic Networks Perspective, Cambridge University Press, 2014; and Diffusion Source Localization in Large Networks, Synthesis Lectures on Communication Networks, Morgan & Claypool Publishers, 2018.
ABSTRACT
The key task in making policy decisions is that of evaluating the impact of various plausible policy choices on the outcomes of interest. For example, what would be the impact of `mask and social distance mandate’ on the health outcomes and economy in Massachusetts in the ongoing pandemic? Causal inference helps answer such “what if scenario analysis” questions. In this tutorial, we will discuss the framework of potential outcomes (Newman ’23, Rubin ’74) viewed from the lens of Tensor Estimation. Through this lens, we will survey the rich body of literature on methodical aspects, mathematical foundations and empirical case studies including the policy design for COVID-19. We will pay special attention to the methods of Synthetic Controls and its recently proposed generalization, Synthetic Interventions. We will provide guidance for empirical practice, with special emphasis on feasibility and data requirements, and characterize the practical settings where methods discussed may be useful and those where they may fail. Finally, we will discuss how the framework of tensor estimation for causal inference is likely to be instrumental in the next wave of development of data efficient reinforcement learning method.
SPEAKER
Devavrat Shah is a Professor in the Department of Electrical Engineering and Computer Science at Massachusetts Institute of Technology. Professor Shah is the founding director of Statistics and Data Science Center at MIT. He is a member of the Institute for Data, Systems and Society, LIDS, CSAIL, and ORC at MIT. His current research interests include algorithms for machine learning, causal inference, and social data processing. He has received paper awards from INFORMS Applied Probability Society, NeurIPS, ACM Sigmetrics, and IEEE Infocom. He has received the Erlang Prize from INFORMS Applied Probability Society and the Rising Star Award from ACM Sigmetrics, and he is a Distinguished Young Alumni of his alma mater, IIT Bombay. In 2013, he founded the machine learning start-up Celect (part of Nike since 2019) which helps retailers with optimizing inventory by accurate demand forecasting.
Day 2: Consensus and Dissensus
ABSTRACT
A long line of work in the mathematical social sciences has considered models of opinion formation in which agents in a network hold opinions drawn from a one-dimensional space, and update their opinions based on different forms of averaging with the opinions of their neighbors. We consider two broad themes in these models: first, the contrast between equilibrium opinions and notions of social optimality; and second, the potential for an adversary to perturb the dynamics with the goal of causing discord. For both of these categories of questions, we show how the underlying behavior is governed by eigenvalues of the network structure. In contrast to typical applications of spectral analysis for networks, where only the extreme eigenvalues determine the outcome, we find here that the entire set of eigenvalues plays an intrinsic role in the dynamics. This talk is based on joint work with Jason Gaitonde and Eva Tardos.
SPEAKER
Jon Kleinberg is the Tisch University Professor in the Departments of Computer Science and Information Science at Cornell University. His research focuses on the interaction of algorithms and networks, the roles they play in large-scale social and information systems, and their broader societal implications. He is a member of the National Academy of Sciences and the National Academy of Engineering, and the recipient of MacArthur, Packard, Simons, Sloan, and Vannevar Bush research fellowships, as well awards including the Harvey Prize, the Nevanlinna Prize, and the ACM Prize in Computing.
ABSTRACT
We introduce a general model of continuous-time opinion dynamics for an arbitrary number of agents that communicate over a network and form real-valued opinions about an arbitrary number of options. Drawing inspiration from existing bio-physical models of neuronal networks and from artificial neural networks, we apply a sigmoidal saturating function to inter-agent and intra-agent exchanges of opinions. The saturating function is the only nonlinearity in the model, yet we prove how it yields consensus, dissensus, and rapid and reliable opinion cascades as a function of just a few parameters. We further show how the network opinion dynamics exhibit both robustness to disturbance and sensitivity to inputs, and we design feedback dynamics for system parameters that enable active tuning of sensitivity thresholds. The general model is being used for systematic study of problems ranging from political polarization and information spreading to spatial decision making for mobile agents. This is joint work with Alessio Franci and Anastasia Bizyaeva.
SPEAKER
Naomi Ehrich Leonard is the Edwin S. Wilsey Professor of Mechanical and Aerospace Engineering at Princeton University. She is a MacArthur Fellow and a Fellow of the American Academy of Arts and Sciences, SIAM, IEEE, IFAC, and ASME. She received her BSE in Mechanical Engineering from Princeton University and her PhD in Electrical Engineering from the University of Maryland. Her research is in control and dynamics with application to multi-agent systems and robotics, collective decision making, spreading processes, mobile sensor networks, and collective animal behavior.
ABSTRACT
Most research on opinion dynamics models focuses on the dynamics of opinions of many individuals on a single topic. Advertising campaigns on the other hand, crucially utilize opinions on multiple topics to shift opinions on a target topic. I will present recent work with Hazla, Jin, and Ramnarayan showing that such campaigns naturally lead to polarization. (Mostly based on https://arxiv.org/abs/1910.05274)
SPEAKER
Elchanan Mossel is a Professor of Mathematics at the Massachusetts Institute of Technology. His research is in the areas of probability, combinatorics, and inference. His interests include combinatorial statistics, discrete Fourier analysis, randomized algorithms, computational complexity, Markov random fields, social choice, game theory, evolution, and the mathematical foundations of deep learning. Mossel received the B.Sc. from The Open University in Israel in 1992 and the M.Sc. (1997) and Ph.D. (2000) degrees in mathematics from the Hebrew University of Jerusalem. He was a postdoctoral fellow at the Microsoft Research Theory Group and a Miller Fellow at UC Berkeley. He joined the UC Berkeley faculty in 2003 where he was a Professor of statistics and computer science. He spent leaves as a Professor at the Weizmann institute (2008-2010) and at the Wharton School, University of Pennsylvania (2014-2016). Mossel is on the senior faculty of the Mathematics Department, with a jointly core faculty appointment at the Statistics and Data Science Center of MIT’s Institute for Data, Systems and Society (IDSS).
Day 3: Epidemic Spreading and Containment
ABSTRACT
Information propagation in networks can be performed using a variety of epidemic mechanisms such as SI, SIS or SIR (susceptible – infective – removed / susceptible). The first part of this talk will address performance of the corresponding epidemic information propagation, with an emphasis on the role played by the network topology. The second part of this talk will consider various mechanisms for orchestrating competing epidemics, as a means to perform real-time data propagation in peer-to-peer networks. Finally, motivated by machine learning applications, we will discuss the use of “gossip-style” information propagation for distributed convex optimization over computer networks.
SPEAKER
Laurent Massoulié is Research Director at Inria, Head of the Microsoft Research – Inria Joint Centre, and Professor at the Applied Maths Centre of Ecole Polytechnique. His research interests are in machine learning, probabilistic modelling and algorithms for networks. He has held research scientist positions at France Telecom, Microsoft Research, and Thomson-Technicolor, where he headed the Paris Research Lab. He received Best Paper awards at IEEE INFOCOM 1999, ACM SIGMETRICS 2005, ACM CoNEXT 2007, and NeurIPS 2018. He was elected “Technicolor Fellow” in 2011, received the “Grand Prix Scientifique” of the Del Ducal Foundation delivered by the French Academy of Science in 2017, and is a Fellow of the “Prairie” Institute.
ABSTRACT
Regional quarantine policies, in which a portion of a population surrounding infections are locked down, are an important tool to contain disease. Further, jurisdictional governments-such as cities, counties, states, and countries-act with minimal coordination across borders. We show that a regional quarantine policy can effectively halt the spread of an infection only if (i) infections have a short enough latency, (ii) a government has essentially complete control over and knowledge of the necessary parts of the network (no leakage of behaviors), and (iii) the network of interactions satisfies a certain balanced-growth condition. With interactions across jurisdictions and nontrivial latency in the detection of infections, regional policies are no longer effective. We show that substantial improvements are possible if jurisdictional governments adopt proactive policies, triggering lockdowns in reaction to neighboring jurisdictions’ infection rates, in some cases even before infections are detected internally. We also show that even a few lax governments, with very myopic or laissez-faire policies, impose substantial costs on the whole system. Our results illustrate the importance of building contagion models that account for contagion across jurisdictions and offer a starting point in designing proactive policies that work across decentralized jurisdictions.
SPEAKER
Matthew O. Jackson is the William D. Eberle Professor of Economics at Stanford University and an External Faculty member of the Santa Fe Institute. His research interests include game theory, microeconomic theory, and the study of social and economic networks, on which he has published many articles and the books The Human Network and Social and Economic Networks. He also teaches an online course on networks and co-teaches two others on game theory. Jackson is a Member of the National Academy of Sciences, a Fellow of the American Academy of Arts and Sciences, a Fellow of the Econometric Society, a Game Theory Society Fellow, and an Economic Theory Fellow. Among his many honors is a Guggenheim Fellowship, the Social Choice and Welfare Prize, the von Neumann Award from Rajk Laszlo College, an honorary doctorate from Aix-Marseille University, the B.E.Press Arrow Prize for Senior Economists, and teaching awards. He is the President of the Game Theory Society.
ABSTRACT
A common theme among many models for spreading processes in networks is the assumption that the propagating object (e.g., a pathogen, in the context of infectious disease propagation, or a piece of information, in the context of information propagation) is transferred across network nodes without going through any modification. However, in real-life spreading processes, pathogens often evolve in response to changing environments or medical interventions, and information is often modified by individuals before being forwarded. In this talk, we will discuss the effects of such adaptations on spreading processes in complex networks with the aim of revealing their role in determining the threshold, probability, and final size of epidemics, and exploring the interplay between them and the structural properties of the network.
SPEAKER
H. Vincent Poor is the Michael Henry Strater University Professor at Princeton University, where he is engaged in research in information theory, machine learning and network science, and their applications in wireless networks, energy systems, and related fields, including recently modeling the spread of the COVID-19 epidemic. He is a member of the National Academy of Engineering and the National Academy of Sciences, and a foreign member of the Chinese Academy of Sciences and the Royal Society. Recognition of his work includes the 2017 IEEE Alexander Graham Medal and honorary doctorates from universities in Asia, Europe, and North America.
Day 4: Information and Disinformation
ABSTRACT
Online social networks are often subject to influence campaigns by malicious actors through the use of automated accounts known as bots. We consider the problem of detecting bots in online social networks and assessing their impact on the opinions of individuals. We begin by developing a bot detection algorithm based on the Ising model from statistical physics. This algorithm can simultaneously identify multiple bots based on their network interaction patterns. Our Ising model algorithm can identify bots with higher accuracy while utilizing much less data than other state of the art methods. We then develop a function we call generalized harmonic influence centrality to estimate the impact bots have on the opinions of users in social networks. This function is based on a generalized opinion dynamics model which subsumes many other models. We prove that this model reaches an equilibrium characterized by a set of linear equations. By combing a neural network for measuring sentiment with the observed social network structure, we are able to calculate the generalized harmonic influence centrality of bots in multiple real social networks involving the 2016 US presidential election, Brexit, and the Gilets Jaunes protests in France. For some networks we find that a limited number of bots can cause non-trivial shifts in the population opinions. In other networks, we find that the bots have little impact. Overall we find that generalized harmonic influence centrality is a useful operational tool to measure the impact of bots in social networks.
SPEAKER
Tauhid Zaman is an Associate Professor of Operations Management at the Yale School of Management. He received his BS, MEng and PhD degrees in electrical engineering and computer science from MIT. His research focuses on solving operational problems involving social network data using probabilistic models, network algorithms, and modern statistical methods. Some of the topics he studies in the social networks space include combating online extremists and assessing the impact of bots. His broader interests cover data driven approaches to investing in startup companies, algorithmic sports betting, and biometric data. His work has been featured in the Wall Street Journal, Wired, the Los Angeles Times, and Time Magazine.
ABSTRACT
A social psychological approach to interpersonal influence systems is described in which a group of individuals is co-oriented to an issue, display their initial positions on the issue, accord relative weights to their own and others’ positions, and alter their displayed positions on the basis of a postulated shared information integration mechanism. The issue may involve attitudes, opinions, beliefs, truth statements, resource allocations, threat assessments, or interpersonal appraisals. The information integration mechanism generates an issue-specific influence network with arcs of accorded weights on which basis the total (direct and indirect) relative influence centralities of the individuals emerge. These centralities influence are metrics of the contributions of each individual’s initial position on the settled positions of each group member. The various applications and experiments that have been published on this model will be described.
SPEAKER
Noah E. Friedkin is a Professor in the Department of Sociology at the University of California, Santa Barbara. He is an AAAS Fellow and Editor-in-Chief of the Journal of Mathematical Sociology. His research interests are in network science modeling of interpersonal influence systems. He has published two award winning books on this subject along with numerous journal publications in which modelling predictions are evaluated with data collected from experiments on human subjects. During the past six years, his work has been conducted with collaborators in the fields of engineering control theory (F. Bullo, R. Tempo, A.V. Proskurnikov) and computer science (A. Singh).
ABSTRACT
This work considers the real-time misinformation detection problem on information spreading networks, where our goal is to detect misinformation quickly and accurately in a scalable way. We formulate this problem as a Markov optimal stopping problem that encodes both the cost from detection error and the cost from letting misinformation spread. Our approach combines model-driven and data-driven methods, where the proposed algorithm, named QuickStop, is an optimal stopping algorithm based on a probabilistic information spreading model obtained from labeled data. The algorithm consists of an offline machine learning algorithm for learning the spreading model and an online algorithm that uses the optimal stopping rule to detect misinformation. The online detection algorithm has both low computational and memory complexities. Our numerical evaluations with a real-world dataset show that QuickStop outperforms existing misinformation detection algorithms in terms of both accuracy and detection time (number of observations needed for detection) and our evaluations with synthetic data show that QuickStop is robust to (offline) learning errors.
SPEAKER
Weina Wang is an Assistant Professor in the Computer Science Department at Carnegie Mellon University. Her research lies in the broad area of applied probability and stochastic systems, with applications in cloud computing, data centers, and privacy-preserving data analytics. From 2016 to 2018, she was a joint postdoctoral research associate in the Coordinated Science Lab at the University of Illinois at Urbana-Champaign and in the School of ECEE at Arizona State University. She received her PhD degree in Electrical Engineering from Arizona State University in 2016 and her Bachelor’s degree from the Department of Electronic Engineering at Tsinghua University in 2009. Her dissertation received the Dean’s Dissertation Award in the Ira A. Fulton Schools of Engineering at Arizona State University in 2016. She received the Kenneth C. Sevcik Outstanding Student Paper Award at ACM SIGMETRICS 2016.