Welcome to the DOS optimization seminar homepage. This seminar series is organized by the H. Milton Stewart School of Industrial and Systems Engineering at Georgia Tech.

DOS is LIVE now! since Fall 2015, we can broadcast all DOS seminars upon the permission of speakers. If you would like to get connected remotely, please send us an email and you will be able to enjoy the seminars from every corner of the world!

Below please find a list of previous and upcoming speakers from 2018 Spring to present.

You can also find a list of archived talks here.

Upcoming Talks

Feb. 22th, 2019 (Friday 11:00 am) - Tauhid Zaman (MIT Sloan School of Management), Optimizing Opinions with Stubborn Agents [Abstract]
Abstract.   Recent news about Russian propaganda bots influencing elections has created a need for social media counter-measures. One approach is to deploy agents in a social network to counter the influence campaign. These can be viewed as "stubborn" agents whose goal is to shift the opinions of non-stubborn people by sharing relevant content. In this talk we show how to deploy stubborn agents in a social network in order to accomplish this. We begin by proposing an opinion dynamics model for interacting individuals in a social network. Our model differs from previous models by allowing individuals to grow more stubborn with time. Despite the time varying nature of our model, we are able to provide a precise characterization of the conditions under which the opinions convergence to an equilibrium. We find that the equilibrium opinions are given by a linear system which has a form similar to Ohm's Law from circuit theory. To validate this opinion model, we use it to predict the opinions of hundreds of thousands of Twitter users discussing varying polarized topics. We use the content of these users tweets as a measure of their true opinions. We develop a neural network to extract the user's opinions from their tweet text. We find that our opinion model accurately predicts the mean opinion of the populations and has a high correlation with their tweet based opinions. Using state of the art detection algorithms, we are able to identify bots in our datasets and use our opinion model to measure the effect they have on the population opinions. We also show how to place stubborn agents in a network to have maximal impact on the population opinion. We formulate this as a discrete optimization problem. We show that the problem is submodular, which allows us to find an accurate and fast solution using a greedy algorithm. We find that a relatively small number of stubborn agents can have a strong effect on the population opinion in several real social networks.

Bio.   Tauhid is an Associate Professor of Operations Management at the MIT Sloan 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 predicting the popularity of content, finding online extremists, and geo-locating users. His broader interests cover data driven approaches to investing in startup companies, non-traditional choice modeling, algorithmic sports betting, and biometric data. His work has been featured in the Wall Street Journal, Wired, Mashable, the LA Times, and Time Magazine.

Previous Talks

Feb. 15th, 2019 (Friday 11:00 am) - Michael Johnson (University of Massachusetts, Boston), Community-Engaged Operations Research: Localized Interventions, Appropriate Methods, Social Impact [Abstract]
Abstract.   Community-engaged operations research is an extension of multiple OR/MS traditions to support participatory research, localized impact and social change. It applies critical thinking, evidence-based policy analysis, community participation and decision modeling to local interventions. It emphasizes the needs, voices and values of disadvantaged and marginalized populations. It rests on a foundation of meaningful engagement with communities. Through a survey of current scholarship in two complementary areas of inquiry, 'community operational research' (referring to work by primarily European researchers) and 'community-based operations research' (referring to work by primarily American researchers), I develop principles for community-engaged OR, present critical questions that represent opportunities to expand the impact of this work, and discuss current projects whose methods and applications have the potential to enrich research and practice in operations research, management science and analytics.

Bio.   https://www.umb.edu/faculty_staff/list/michael_johnson
Oct. 26th, 2018 (Friday 11:15 am) - Thinh T. Doan, Distributed learning over a network of agents under communication constraints! [Abstract]
Abstract.   In this talk, I will present a popular distributed method, namely, distributed consensus-based gradient (DCG) method, for solving optimal learning problems over a network of agents. Such problems arise in many applications such as, finding optimal parameters over a large dataset distributed among a network of processors or seeking an optimal policy for coverage control problems in robotic networks. The focus is to present our recent results, where we study the performance of DCG when the agents are only allowed to exchange their quantized values due to their finite communication bandwidth. In particular, we develop a novel distributed variant of the popular stochastic approximation under random quantization. Our notable contribution is to provide an explicit formula for their rates of convergence, which shows the dependence on the underlying network topology and the size of the bandwidths shared between agents. Finally, I conclude my talk with some discussion about potential applications of distributed stochastic approximations in multi-agent reinforcement learning.

Bio.   Thinh T. Doan is a TRIAD postdoctoral fellow at Georgia Institute of Technology, joint between the School of Industrial and Systems Engineering and the School of Electrical and Computer Engineering (ECE). He was born in Vietnam, where he got his Bachelor degree in Automatic Control at Hanoi University of Science and Technology in 2008. He obtained his Master and Ph.D. degrees both in ECE from the University of Oklahoma in 2013 and the University of Illinois at Urbana-Champaign in 2018, respectively. His research interests lie at the intersection of control theory, optimization, distributed algorithms, and applied probability, with the main applications in machine learning, reinforcement learning, power networks, and multi-agent systems.
Oct. 19th, 2018 (Friday 11:15 am) - Sebastian Salazar, Dynamic Resource Allocation in the Cloud with Near-Optimal Efficiency [Abstract]
Abstract.   The new era of Cloud Computing has given average users the benefit of high performance technology without specialized infrastructure or in-depth knowledge. Sharing resources efficiently between different users, and taking advantage of economies of scale make this new business model economically viable. In many cloud settings, providers would like to operate resources at high utilization while simultaneously satisfying user requirements. Nonetheless, there is an unavoidable trade-off between these two objectives. For instance, focusing solely on satisfying individual requirements can lead to poor utilization. In this talk, I will introduce a multiplicative weights algorithm for dynamic allocation of a single divisible resource, such as CPU, that guarantees simultaneously near-optimal utilization and user satisfaction. Joint work with Ishai Menache, Mohit Singh and Alejandro Toriello.

Bio.   Sebastian is a second-year PhD student in ACO, currently working with Mohit Singh and Alejandro Toriello on optimization, online algorithms and approximation algorithms.
Oct. 12th, 2018 (Friday 11:15 am) - Gerdus Benade(CMU), How to make envy vanish over time [Abstract]
Abstract.   We study the dynamic fair division of indivisible goods. Suppose at every time step an item arrives online and must be allocated to an agent upon arrival, and that agents have heterogeneous values for the items. Our goal is to design allocation algorithms that minimize the maximum envy, defined as the maximum difference between any agent's overall value for items allocated to another agent and to herself. We say that an algorithm has vanishing envy if the ratio of envy over time goes to zero as time goes to infinity. We design a polynomial-time, deterministic algorithm that achieves vanishing envy, and show the rate at which envy vanishes is asymptotically optimal. We also derive tight (in the number of items) bounds for a more general setting where items arrive in batches.

Bio.   Gerdus Benade is a Ph.D. student in Operations Research in the Tepper School of Business at Carnegie Mellon University. He is interested in discrete optimization and computational social choice.
Oct. 5th, 2018 (Friday 11:15 am) - Swati Gupta, Mathematical Theories of Bias and Fairness [Abstract]
Abstract.   With ubiquitous automated decision-making comes great responsibility (or does it?). Modern optimization methods shape and affect almost all aspects of modern life -- search, finance, urban transportation, credit applications, and criminal justice, to name a few. It is important that we understand (and mitigate) the unintended consequences of automated optimized decisions to avoid discrimination among the users (even perceived discrimination), as well as ensure due process and transparency in decision-making. For instance, (i) delivery of certain services has been made available at a higher cost based on the location of the customers thus inadvertently discriminating against minority neighborhoods, (ii) algorithms like COMPAS used in the criminal justice system have been shown to incorrectly classify black defendants as high risk with much higher likelihood than white defendants, and (iii) experiments on recommendation engines have shown that women are much less likely to be shown higher paying jobs compared to men. These are just some examples in which automated decisions have had adverse effects on various members of the population. In this talk, we will survey some of the mathematical theories developed to model bias and fairness from a statistical as well as an optimization perspective. We will discuss impossibility results and trade-offs between various fairness metrics proposed in the literature, with the hope of informing policymakers on what is even achievable mathematically. We will also discuss some on-going work along these lines in the context of making balanced inventory routing decisions, placing public facilities while ensuring equity and matching students fairly to schools, as well as interesting open questions in this domain.

Bio.   Dr. Swati Gupta is an Assistant Professor in the H. Milton Stewart School of Industrial and Systems Engineering at Georgia Tech. Gupta's research interests lie primarily in combinatorial, convex, and robust optimization with applications in online learning and data-driven decision-making under partial information. Her work focuses on speeding up fundamental bottlenecks that arise in learning problems due to the combinatorial nature of the decisions, as well as drawing from machine learning to improve traditional optimization methods.
Sep. 28th, 2018 (Friday 11:15 am) - Ashia C. Wilson (Microsoft Research), A Dynamical View of Optimization Algorithms [Abstract]
Abstract.   Optimization is a core primitive in statistics, machine learning, and data analytics. Within these fields, the rapid growth in the scale and complexity of modern datasets has led to a focus on two classes of algorithms: gradient methods and momentum methods (also referred to as accelerated methods). Momentum methods, first proposed by Nesterov in 1983, achieve faster convergence rates than gradient methods. However, unlike gradient methods, they are not descent methods and providing robust performance guarantees remains a challenge. In simple settings, momentum methods can be understood as modeling the dynamics of a damped harmonic oscillator; making this intuition precise however, and generalizing it to other geometries has been difficult. Furthermore, derivations of momentum methods do not flow from a single underlying principle, but tend to rely on case-specific algebra using a technique - considered by many to be esoteric - called estimate sequences. The first part of our work introduces a variational, continuous-time framework for understanding momentum methods. We show that there is a family of Lagrangian functionals, which we call Bregman Lagrangians, which generate dynamics corresponding to momentum methods in continuous time. In particular, momentum methods can be understood as arising from various discretization techniques applied to these continuous time dynamics. The second part of our work strengthens this connection. We demonstrate how to derive families of Lyapunov functions which can certify rates of convergence for the continuous time momentum dynamics. We further demonstrate how the proofs of convergence of momentum methods can be understood as bounding discretization errors of the Lyapunov function when moving to discrete time. Along the way, we prove an equivalence between these family of Lyapunov functions and the technique of estimate sequences as well as demonstrate how this framework allows us to derive and analyze several new optimization methods. The following is joint work with Andre Wibisono, Stephen Tu, Shivaram Venkataraman, Alex Gittens, Benjamin Recht, and Michael I. Jordan.

Sep. 21st, 2018 (Friday 11:15 am) - Natashia Boland , Decomposition Branching for Mixed Integer Programming [Abstract]
Abstract.   In the last few decades, there have been enormous advances in exact methods for the solution of general mixed integer programming (MIP) problems. These advances have come from attention to several key aspects of the LP-based branch-and-bound framework, including preprocessing, cutting planes, search strategies, and branching variable selection. However, the branching constraints themselves have received relatively little attention, and those used in modern codes remain the same as they were in the 1960's. This talk explores one idea for branching that seeks to exploit decomposable structure in MIPs: decomposition branching. The approach uses the power of MIP solvers to solve (or nearly solve) smaller subproblems so as to derive the branching constraint. The computational cost of doing so, in experiments with instances of a weighted set covering problem and a regionalized p-median facility location problem with assignment range constraints, is greatly outweighed by its benefits. The number of branch-and-bound nodes is vastly reduced and run times can be orders of magnitude less than those of a commercial solver.

Bio.   Natashia Boland is a professor in the Stewart School of Industrial and Systems Engineering at Georgia Tech. Recently, her work has focused on two main directions. The first, a better solution of planning problems in time-space networks: she has developed techniques that generate only the parts of the network really needed to determine near-optimal solutions, with high accuracy and in far faster computing times than previously available. She is applying these concepts in the design and operation of less-than-truckload carrier networks. The second is on addressing problems with multiple objectives. She is developing technology that can efficiently, in practice, offer the decision maker a comprehensive view of the trade-offs available to them, given the constraints of their system. Such technology is especially helpful in settings in which issues such as fairness, risk, environmental impact, and human factors, as well as costs or profits, play an important role in the decision-making.
Sep. 14th, 2018 (Friday 11:15 am) - Pascal Van Hentenryck , Hybrid Optimization: The Many Marriages of CP, MIP, and SAT [Abstract]
Abstract.   In the last decades, significant progress in discrete optimization originated from hybridizations of synergic optimization methods. This talk presents three case studies of combining constraint programming, mixed-integer programming, and satisfiability. The benefits are illustrated on applications in routing and scheduling.

Bio.   Pascal Van Hentenryck is the A. Russell Chandler III Chair and Professor in the H. Milton Stewart School of Industrial and Systems Engineering at Georgia Tech. His current research focuses on artificial intelligence, data science, and operations research with applications in energy, mobility, and privacy.
Apr. 6th, 2018 (Friday 11am) - Rediet Abebe (Cornell), Information and Overexposure in Word-of-Mouth Campaigns [Abstract]
Abstract.   Access to information is of major concern in a wide range of domains including health, education, and labor markets. Word-of-mouth campaigns are an effective tool for disseminating information that can be leveraged to improve people's lives. Within health, specifically, by targeting individuals with crucial health information that might be relevant to them, we can potentially create an information cascade that reaches a large portion of the population, and in turn, improve health outcomes for many people.

In traditional models for word-of-mouth campaigns, the objective has generally been based on reaching as many people as possible. However, a number of studies both within health and in commercial settings have shown that the indiscriminate spread of a product by word-of-mouth can result in overexposure where the information or product reaches a part of the population that is not receptive to it. This can lead to negative reputational effects or decrease the efficacy of future campaigns. In the first part of the talk, we ask how we should make use of social influence when there is a risk of overexposure. We develop and analyze a theoretical model for this process; we show that it captures a number of the qualitative phenomena associated with overexposure and provide an efficient algorithm to find the optimal campaigning strategy.

In the second part of the talk, we focus on the stark inequalities in the availability of health-related data. We focus on the lack of data on health information needs of individuals in Africa and explore the role that search engines can play in uncovering people's everyday health information needs, concerns, and misconceptions. We analyze Bing searches related to HIV/AIDS in all 54 nations in Africa, and show that these expose a wide range of themes that shed light on how health information needs vary geographically as well as by demographic groups. We discuss the potential for these results to inform targeted education efforts to decrease the burden of the disease.

Bio.   Rediet Abebe is a Ph.D. candidate in Computer Science at Cornell. Her research focuses on algorithms, artificial intelligence, and applications to social good. Her work has generously been supported by fellowships and scholarships through Facebook (2017-2019), Google (2016-2017), and the Cornell Graduate School (2015-2016). She is also a 2013-2014 Harvard Cambridge Scholar.
Apr. 10th, 2018 (Tuesday 11am) - Weijun Xie (VT), Multi-Product Newsvendor Model with Substitutions [Abstract]
Abstract.   We study a multi-product newsvendor model with customer-driven demand substitutions, where each product, once run out of stock, can be proportionally substituted by the others. This model has been widely studied in the literature, however, due to its nonconvexity and intractability, only very limited analytical properties have been discovered nor efficient solution approaches. This paper first completely characterizes the optimal order policy when the demand is known and recasts the model as a discrete submodular maximization problem. When the demand is stochastic, we first reformulate this nonconvex model as a two-stage stochastic integer program, where the recourse decision variables are mixed integers. We further study a tight upper bound via nonanticipativity dual, which is proven to be close to the true optimal and can be computed as a linear program. Finally, numerical studies demonstrate the effectiveness of the algorithms.

Bio.   Dr. Weijun Xie graduated from Georgia Tech in 2017 and is Assistant Professor in the Department of Industrial and Systems Engineering at Virginia Tech.
Mar. 30th, 2018 (Friday 11am) - Gauri Joshi (CMU), Slow and Stale Gradients Can Win the Race: Error-Runtime Trade-offs in Distributed SGD [Abstract]
Abstract.   Distributed Stochastic Gradient Descent (SGD) when running in a synchronous manner, suffers from delays in waiting for the slowest learners (stragglers). Asynchronous methods can alleviate stragglers, but cause gradient staleness that can adversely affect convergence. In this work, we present the first theoretical characterization of the speed-up offered by asynchronous methods by analyzing the trade-off between the error in the trained model and the actual training runtime (wallclock time). The novelty in our work is that our runtime analysis considers random straggler delays, which helps us design and compare distributed SGD algorithms that strike a balance between stragglers and staleness. We also present a new convergence analysis of asynchronous SGD variants without bounded or exponential delay assumptions, and a novel learning rate schedule to compensate for gradient staleness.

Bio.   Gauri Joshi is an assistant professor in the ECE department at Carnegie Mellon University since September 2017. Prior to that, she worked as a Research Staff Member at IBM T. J. Watson Research Center. Gauri completed her Ph.D. from MIT EECS in June 2016. She also received her B.Tech and M. Tech in Electrical Engineering from the Indian Institute of Technology (IIT) Bombay in 2010. Her awards and honors include the Best Thesis Prize in Computer science at MIT (2012), Institute Gold Medal of IIT Bombay (2010), Claude Shannon Research Assistantship (2015-16), and the Schlumberger Faculty for the Future fellowship (2011-2015).
Feb. 13th, 2018 (Tuesday 11am) - Roy Schwartz (Technion), Fast and Simple Algorithms for Submodular Maximization [Abstract]
Abstract.   Submodular maximization captures both classical problems in combinatorial optimization and practical applications that arise in other disciplines, e.g., machine learning and data mining. The size of the inputs in these applications is usually very large. Hence, it is interesting to devise approximation algorithms that in addition to providing a provable guarantee are also very fast and simple to use. In this talk, I will present several such examples from recent years.

Bio.   Roy Schwartz is an assistant professor at the Computer Science Department at the Technion. His research focuses on the design and analysis of algorithms, approximation algorithms, combinatorial optimization, the geometry of metric spaces and its applications, submodular optimization, and randomized algorithms.
Jan. 23rd, 2018 (Tuesday 11am) - David Hemmi (Monash), High-level modeling and solution of combinatorial stochastic programs [Abstract]
Abstract.   Stochastic programming is concerned with decision making under uncertainty, seeking an optimal policy with respect to a set of possible future scenarios. While the value of Stochastic Programming is obvious to many practitioners, in reality, uncertainty in decision making is oftentimes neglected. For deterministic optimisation problems, a coherent chain of modeling and solving exists. Employing standard modeling languages and solvers for stochastic programs is however difficult. First, they have (with exceptions) no native support to formulate Stochastic Programs. Secondly solving stochastic programs with standard solvers (e.g. MIP solvers) is often computationally intractable. I will talk about my research that aims to make Stochastic Programming more accessible. First, I will discuss modeling deterministic and stochastic programs in the Constraint Programming language MiniZinc - a modeling paradigm that retains the structure of a problem much more strongly than MIP formulations. Second, I will talk about decomposition algorithms I have been working on to solve combinatorial Stochastic Programs.

Bio.   David is a PhD student at Monash University in Melbourne. Before joining Monash, he completed a Bachelors degree in Systems Engineering in Switzerland, studied Robotics at University of Bristol and worked in various engineering roles. As part of his Phd, David is working on a solver system for stochastic combinatorial optimisation problems that are modeled in in the Constraint Programming language MiniZinc. In future, he would like to combine his passion for automation with the research he is doing in operations research.
*Here you can find a list of the archived talks.

DOS History

The DOS Optimization Seminars began in Fall 2007 as the Discrete Optimization Seminars (DOS). These seminars were meant to extend the scope of the Integer Programming Seminars organized by Marcos Goycoolea and Ricardo Fukasawa in previous semesters.

The scope of the Discrete Optimization Seminars was defined as "discrete optimization in its broadest sense" (i.e. integer and combinatorial optimization and related areas such as network flows, stochastic, continuous and global optimization). The extension of the seminars to these related areas lead to the elimination of the "Discrete" qualifier from the seminars title in Spring 2008. The acronym "DOS" was kept simply to keep the web pages theme. The current explanation for DOS is simply a recursive acronym for "DOS Optimization Seminars" in the spirit of the GNU acronym. This is the old logo of the seminars: