Past Computational Sustainability Virtual Seminar Series
The Computational Sustainability Virtual Seminar Series will present talks by researchers and educators in Computational Sustainability, and is being sponsored by CompSustNet, with support from the National Science Foundation's Expeditions in Computing program.
To sign up for seminar announcements, send an email to email@example.com with the word join as the subject (leave the message body empty).
For current seminar information, please see the Spring 2018 listing.
See also the CompSust Open Graduate Seminar (COGS).
|Tue Sep 27, 2016, 4-5pm EDT (UTC-4)||Stefano Ermon, Stanford University||Measuring progress towards sustainable development goals with machine learning|
|Tue Oct 11, 2016, 4-5pm EDT (UTC-4)||Thomas Dietterich, Oregon State University||Solving MDPs for Ecosystem Management: Lessons Learned|
|Tue Oct 25, 2016, 4-5pm EDT (UTC-4)||Milind Tambe and Eric Rice, University of Southern California||How Can AI be Used for Social Good? Key Techniques, Applications, and Results|
|Tue Nov 8, 2016, 4-5pm EST (UTC-5)||Bistra Dilkina, Georgia Institute of Technology||Network Design Approaches to Multi-species Biodiversity Conservation|
|Tue Nov 29, 2016, 4-5pm EST (UTC-5)||Warren B. Powell, Princeton University||A Unified Framework for Handling Decisions and Uncertainty In Energy and Sustainability|
|Thu Dec 15, 2016, 4-5pm EST (UTC-5)||Daniel Sheldon, University of Massachusetts Amherst and Mount Holyoke College||Advances in Probabilistic Inference and Machine Learning for Ecosystem Monitoring|
|Fri Mar 3, 2017, 1:30-2:30pm EST (UTC-5)||Zico Kolter, Carnegie Mellon University||Task-based end-to-end learning in stochastic optimization|
|Fri Mar 17, 2017, 1:30-2:30pm EDT (UTC-4)||David Shmoys, Cornell University||Models and Algorithms for the Operation and Design of Bike-sharing Systems|
|Fri Apr 28, 2017, 1:30-2:30pm EDT (UTC-4)||Mary Lou Zeeman, Bowdoin College||Modeling Challenges in the Food System|
|Fri May 5, 2017, 1:30-2:30pm EDT (UTC-4)||Fei Fang, Harvard University||Game-Theoretic Approaches for Sustainability Challenges|
|Fri May 19, 2017, 1:30-2:30pm EDT (UTC-4)||Angela Fuller, U.S. Geological Survey, Cornell University||Density-Weighted Connectivity for Landscape Management and Connectivity Conservation|
|Fri June 2, 2017, 1:30-2:30pm EDT (UTC-4)||Doug Fisher, Vanderbilt University||A survey of CompSustNet research, education, and outreach|
|Fri Jun 16, 2017, 1:30-2:30pm EDT (UTC-4)||Alan Fern, Oregon State University||Bringing Bayesian Optimization into the Lab: Reasoning about Resources and Actions|
|Fri Oct 20, 2017, 1:30-2:30pm EDT (UTC-4)||Kevin Leyton-Brown, University of British Columbia||Kudu: An Electronic Market for Agricultural Trade in Uganda|
|Fri Nov 3, 2017, 1:30-2:30pm EDT (UTC-4)||Paul L. Fackler, North Carolina State University||Solving stochastic dynamic programming models without transition matrices|
|Fri Nov 17, 2017, 1:30-2:30pm EST (UTC-5)||Kristian Kersting, Technical University Darmstadt||Feeding the World with Big Data: Uncovering Spectral Characteristics and Dynamics of Stressed Plants|
|Fri Dec 1, 2017, 1:30-2:30pm EST (UTC-5)||Michela Milano, University of Bologna||Empirical Model Learning for Sustainability Challenges|
|Fri Dec 15, 2017, 1:30-2:30pm EST (UTC-5)||Nicole Sintov, The Ohio State University||Theory + practice: Promoting sustainable behavior in real-world settings|
2016 Fall Seminar Details
Stefano Ermon, Stanford University
Tue Sep 27, 2016, 4-5pm EDT (UTC-4)
Title: Measuring progress towards sustainable development goals with machine learning
Abstract: Recent technological developments are creating new spatio-temporal data streams that contain a wealth of information relevant to sustainable development goals. Modern AI techniques have the potential to yield accurate, inexpensive, and highly scalable models to inform research and policy. As a first example, I will present a machine learning method we developed to predict and map poverty in developing countries. Our method can reliably predict economic well-being using only high-resolution satellite imagery. Because images are passively collected in every corner of the world, our method can provide timely and accurate measurements in a very scalable end economic way, and could revolutionize efforts towards global poverty eradication. As a second example, I will present some ongoing work on monitoring agricultural and food security outcomes from space.
Bio: Stefano Ermon is currently an Assistant Professor in the Department of Computer Science at Stanford University, where he is affiliated with the Artificial Intelligence Laboratory and the Woods Institute for the Environment. He completed his PhD in computer science at Cornell in 2015. His research interests include techniques for scalable and accurate inference in graphical models, statistical modeling of data, large-scale combinatorial optimization, and robust decision making under uncertainty, and is motivated by a range of applications, in particular ones in the emerging field of computational sustainability. Stefano has won several awards, including two Best Student Paper Awards, one Runner-Up Prize, and a McMullen Fellowship.
Thomas Dietterich, Oregon State University
Tue Oct 11, 2016, 4-5pm EDT (UTC-4)
Title: Solving MDPs for Ecosystem Management: Lessons Learned
Abstract: Many ecosystem management problems can be formulated as MDPs in which the transition dynamics of the ecosystem is defined by a simulator (as opposed to a probability transition matrix). We have been studying two ecosystem management problems: wildfire management in ponderosa pine forests and tamarisk invasion of river networks. This talk will describe the methods we have developed and the lessons we have learned while working on these problems. For small simulator-defined MDPs, we have developed efficient algorithms that provide probabilistic (PAC) accuracy guarantees. For larger problems, we have found success using Bayesian optimization methods to search a parameterized policy space. The talk will conclude with a set of open research questions.
Bio: Dr. Dietterich (AB Oberlin College 1977; MS University of Illinois 1979; PhD Stanford University 1984) is Professor Emeritus and Director of Intelligent Systems Research in the School of Electrical Engineering and Computer Science at Oregon State University, where he joined the faculty in 1985. Dietterich is one of the pioneers of the field of Machine Learning and has authored more than 130 refereed publications and two books. His research is motivated by challenging real world problems with a special focus on ecological science, ecosystem management, and sustainable development. He is best known for his work on ensemble methods in machine learning including the development of error-correcting output coding. Dietterich has also invented important reinforcement learning algorithms including the MAXQ method for hierarchical reinforcement learning.
Milind Tambe and Eric Rice, University of Southern California
Tue Oct 25, 2016, 4-5pm EDT (UTC-4)
Title: How Can AI be Used for Social Good? Key Techniques, Applications, and Results
Abstract: Discussions about the future negative consequences of AI sometimes drown out discussions of the current accomplishments and future potential of AI in helping us solve complex societal problems. At the USC Center for AI in Society, CAIS, our focus is on exploring AI research in tackling wicked problems in society. This talk will highlight the goals of CAIS and three areas of ongoing work. First, we will focus on the use of AI for assisting low-resource sections of society, such as homeless youth. Harnessing the social networks of such youth, we will illustrate the use of AI algorithms to help more effectively spread health information, such as for reducing risk of HIV infections. These algorithms have been piloted in homeless shelters in Los Angeles, and have shown significant improvements over traditional methods. Second, we will outline the use of AI for protection of forests, fish, and wildlife; learning models of adversary behavior allows us to predict poaching activities and plan effective patrols to deter them. These algorithms are in use in multiple countries, and we discuss concrete results we have obtained in a national park in Uganda. Finally, we will focus on the challenge of AI for public safety and security, discussing game theoretic algorithms for effective security resource allocation that are in actual daily use by agencies such as the US Coast Guard and the Federal Air Marshals Service to assist the protection of ports, airports, flights, and other critical infrastructure. These are just a few of the projects at CAIS, and we expect these and future projects at CAIS to continue to illustrate the significant potential that AI has for social good.
Bio: Milind Tambe is Founding Co-Director of CAIS, the USC Center for AI for Society, and Helen N. and Emmett H. Jones Professor in Engineering at the University of Southern California(USC). He is a fellow of AAAI and ACM, as well as recipient of the ACM/SIGART Autonomous Agents Research Award, Christopher Columbus Fellowship Foundation Homeland security award, INFORMS Wagner prize for excellence in Operations Research practice, Rist Prize of the Military Operations Research Society, IBM Faculty Award, Okawa foundation faculty research award, RoboCup scientific challenge award, and other local awards such as the Orange County Engineering Council Outstanding Project Achievement Award, USC Associates award for creativity in research and USC Viterbi use-inspired research award. Prof. Tambe has contributed several foundational papers in AI in areas such as multiagent teamwork, distributed constraint optimization (DCOP) and security games. For this research, he has received the "influential paper award" and a number of best paper awards at conferences such as AAMAS, IJCAI, IAAI and IVA. In addition, Prof. Tambe pioneering real-world deployments of "security games" has led him and his team to receive the US Coast Guard Meritorious Team Commendation from the Commandant, US Coast Guard First District's Operational Excellence Award, Certificate of Appreciation from the US Federal Air Marshals Service and special commendation given by LA Airport police from the city of Los Angeles. For his teaching and service, Prof. Tambe has received the USC Steven B. Sample Teaching and Mentoring award and the ACM recognition of service award. He has also co-founded a company based on his research, Avata Intelligence, where he serves as the director of research. Prof. Tambe received his Ph.D. from the School of Computer Science at Carnegie Mellon University.
Bistra Dilkina, Georgia Institute of Technology
Tue Nov 8, 2016, 4-5pm EST (UTC-5)
Title: Network Design Approaches to Multi-species Biodiversity Conservation
Abstract: Curbing biodiversity loss is one of the key goals in achieving sustainable development. However, most conservation investments are done with limited budget, and in the face complex spatial variations in economic costs and ecological benefits. I address several hard spatial optimization problems that arise in the context of conservation planning, and show how network design and mixed-integer optimization can be leveraged for finding solutions and supporting effective and cost-efficient decision making. I will present a computational framework for finding optimal wildlife corridors serving multiple species. Our framework enables the systematically study of tradeoffs between economic costs and conservation benefits, tradeoffs between single-species and multi-species planning, as well as tradeoffs with respect to species prioritization. We apply our approach in western Montana to the conservation of grizzly bears and wolverines, and demonstrate economies of scale and complementarities conservation planners can achieve by optimizing corridor designs for financial costs and for multiple species connectivity jointly.
Bio: Bistra Dilkina is an assistant professor in the College of Computing at the Georgia Institute of Technology. She received her PhD from Cornell University in 2012, and was a Post-Doctoral associate at the Institute for Computational Sustainability until 2013. Her research focuses on advancing the state of the art in combinatorial optimization techniques for solving real-world large-scale problems, particularly ones that arise in sustainability areas such as biodiversity conservation planning and urban planning. Her work spans discrete optimization, network design, stochastic optimization, and machine learning. She is also the co-director of the Data Science for Social Good (DSSG) Atlanta summer program, which partners student teams with government and nonprofit organizations to help solve some of their most difficult problems using analytics, modeling, prediction and visualization.
Warren B. Powell, Princeton University
Tue Nov 29, 2016, 4-5pm EST (UTC-5)
Title: A Unified Framework for Handling Decisions and Uncertainty In Energy and Sustainability
Abstract: Problems in energy and sustainability represent a rich mixture of decisions intermingled with different forms of uncertainty. These decision problems have been addressed by multiple communities from operations research (stochastic programming, Markov decision processes, simulation optimization, decision analysis), computer science, optimal control (from engineering and economics), and applied mathematics. In this talk, I will identify the major dimensions of this rich class of problems, spanning static to fully sequential problems, offline and online learning (including so-called "bandit" problems), derivative-free and derivative-based algorithms, with attention given to problems with expensive function evaluations. We divide solution strategies for sequential problems ("dynamic programs") between stochastic search ("policy search") and policies based on lookahead approximations (which include both stochastic programming as well as value functions based on Bellman's equations). We further divide each of these two fundamental solution approaches into two subclasses, producing four classes of policies for approaching sequential stochastic optimization problems. We use a simple energy storage problem to demonstrate that each of these four classes may work best, as well as opening the door to a range of hybrid policies. I will show that a single elegant framework spans all of these approaches, providing scientists with a more comprehensive toolbox for approaching the rich problems that arise in energy and sustainability.
Bio: Warren B. Powell is a professor in the Department of Operations Research and Financial Engineering at Princeton University, where he has taught since 1981 after receiving his BSE from Princeton University and Ph.D. from MIT. He is the founder and director of the laboratory for Computational Stochastic Optimization and Learning (CASTLE Labs), which spans contributions to models and algorithms in stochastic optimization, with applications to energy systems, health and medical research, and the sciences. He has two books and over 200 papers, and is working on a new book "Optimization under Uncertainty: A Unified Framework."
Daniel Sheldon, University of Massachusetts Amherst and Mount Holyoke College
Thu Dec 15, 2016, 4-5pm EST (UTC-5)
Title: Advances in Probabilistic Inference and Machine Learning for Ecosystem Monitoring
Abstract: Machine learning combined with large and novel data resources can contribute to our understanding of ecosystems in a variety of ways. This talk will describe two different applications of machine learning to ecosystem monitoring. First, I will describe our ongoing work to measure continent-scale bird migration using archived weather radar data. Machine learning algorithms automate the complex process of interpreting radar imagery and allow us to access high-level biological information in this massive data archive. Second, I will describe advances in probabilistic inference for estimating animal population parameters from survey data. We present the first exact polynomial-time inference algorithms for a class of commonly used models that include latent count variables to represent unknown population sizes. Our approach uses probability generating functions to represent and manipulate the infinite sequences that one must reason about during inference, and is much faster than existing approximate approaches.
Bio: Daniel Sheldon is an Assistant Professor of Computer Science at the University of Massachusetts Amherst and Mount Holyoke College. He received his Ph.D. from the Department of Computer Science at Cornell University in 2009, and was an NSF Postdoctoral Fellow in Bioinformatics at the School of EECS at Oregon State University from 2010-2012. His research interests are in machine learning, probabilistic modeling, and optimization applied to large-scale problems in ecology, computational sustainability, and networks. His work was recognized by a Computational Sustainability Best Paper Award at AAAI 2013, and is supported by the NSF and MassDOT.
2017 Spring Seminar Details
Zico Kolter, Carnegie Mellon University
Fri Mar 3, 2017, 1:30-2:30pm EST (UTC-5)
Title: Task-based end-to-end learning in stochastic optimization
Abstract: In this talk, I will present recent work in learning predictive models for use in stochastic optimization settings. In these domains, the goal of a probabilistic model is not merely to generate "accurate" predictions of the future, but also to make predictions that will result in an effective policy when integrated into decision-making processes. These two goals may seem well-aligned, but they often differ to a surprising degree when models do not reflect the true underlying system (the norm rather than the exception in machine learning). To address this challenge, we develop a technique we refer to as task-based end-to-end learning. The main idea is to update the predictive model itself through the task-specific loss, by differentiating through the policy decisions, in order to directly improve the policy performance under the true distribution of interest. This in turn requires techniques for differentiating through the solution of general optimization problems, a task for which we develop the algorithms and an efficient implementation. We apply this method to the task of scheduling generation under uncertain demand and ramping constraints, and shows that it can significantly outperform a naive maximum likelihood approach.
Bio: Zico Kolter is an Assistant Professor in the School of Computer Science at Carnegie Mellon University, with appointments in the Computer Science Department, the Institute for Software Research (in the Societal Computing program), and affiliated appointments with the Machine Learning Department, the Robotics Institute, and the Electrical and Computer Engineering Department. His work focuses on machine learning and optimization, with a specific focus on applications in smart energy systems. From an algorithmic standpoint, he has worked on fast optimization algorithms for a number of problems and for general convex programs, large-scale probabilistic modeling, stochastic optimization, and deep learning. On the application side, he has worked on energy disaggregation, probabilistic forecasting for energy systems, and model predictive control techniques for industrial control in the electrical grid.
David Shmoys, Cornell University
Fri Mar 17, 2017, 1:30-2:30pm EDT (UTC-4)
Title: Models and Algorithms for the Operation and Design of Bike-sharing Systems
Abstract: The sharing economy has helped to transform many aspects of our day-to-day lives, leveraging the IT revolution in increasingly novel ways. At the same time, the sharing economy presents new computational challenges to provide tools to support the operations of these emerging industries. Although perhaps not quite as visible in impact as Uber and Airbnb (and their competitors), bike-sharing systems have fundamentally changed the urban landscape as well. Even in a city as notoriously inhospitable to cycling as New York, Citibike has emerged as a significant player in the city's transportation network, supporting more than 1.5 million rides per month for a subscriber base of more than 100,000 individuals. We have been working with Citibike to develop analytics and optimization models and algorithms to help manage this system. The key challenge is to cope with huge rush-hour usage that simultaneously creates stark shortages of bikes in some neighborhoods, and surpluses of bikes (and consequently, shortages of parking docks) elsewhere. We will explain how mathematical models can be used to answer questions such as, how should we position the fleet of bikes at the start of a day, and how should we mitigate the imbalances that develop? We will survey the analytics we have employed for the former question, where we developed an approach based on continuous-time Markov chains combined with optimization models to compute daily stocking levels for the bikes, as well as methods employed for optimizing the capacity of the stations. For the question of mitigating the imbalances that result, we will describe algorithms that guide both mid-rush hour and overnight rebalancing, as well as for the positioning of corrals, which create "surge capacity" at stations, and have been one of the most effective means of creating adaptive capacity in the system.
This is a survey of several papers, but will focus on joint work with Daniel Freund, Shane Henderson, and Eoin O'Mahony.
Bio: David Shmoys is the Laibe/Acheson Professor at Cornell University in the School of Operations Research and Information Engineering, and also the Department of Computer Science at Cornell University, and is currently the Director of the School of Operations Research and Information Engineering. Shmoys's research has focused on the design and analysis of efficient algorithms for discrete optimization problems, with applications including scheduling, inventory theory, computational biology, computational sustainability, and most recently, on optimization models and algorithms for issues underlying the sharing economy. His graduate-level text, The Design of Approximation Algorithms, co-authored with David Williamson, was awarded the 2013 INFORMS Lanchester Prize. He is a Fellow of INFORMS, ACM, and SIAM. He has served on numerous editorial boards, having served as Editor-in-Chief of SIAM J. on Discrete Math, and Research in the Mathematical Sciences (for theoretical computer science), and is currently an Associate Editor of Mathematics of Operations Research.
Mary Lou Zeeman, Bowdoin College
Fri Apr 28, 2017, 1:30-2:30pm EDT (UTC-4)
Title: Modeling Challenges in the Food System
Abstract: We will describe some challenges in food systems and in resilience that might lend themselves to a collaborative approach between computer scientists, mathematicians and domain scientists. We will also briefly discuss some of the mechanisms that the Mathematics and Climate Research Network has found helpful in creating and sustaining geographically dispersed online working groups and mentoring teams.
Bio: Dr. Mary Lou Zeeman uses mathematics within cross-disciplinary research communities to help understand sustainability, climate change, and protecting the health of the planet. Zeeman is a co-director of the Mathematics and Climate Research Network, a member of the executive council of the Computational Sustainability Network, and a co-leader of the Mathematics of Planet Earth Initiative. She is a professor of Mathematics at Bowdoin College.
Fei Fang, Harvard University
Fri May 5, 2017, 1:30-2:30pm EDT (UTC-4)
Title: Game-Theoretic Approaches for Sustainability Challenges
Abstract: The framework of game theory can be powerful when addressing resource allocation problems in security and sustainability domains, e.g., protecting critical infrastructure and cyber network, and protecting wildlife, fishery, and forest. Motivated by these problems, I propose models and algorithms to handle massive games with complex spatio-temporal settings, leading to real-world applications that have fundamentally altered current practices of security resource allocation. In this talk, I will focus on my work motivated by environmental sustainability challenges. First, for problems with repeated interaction such as preventing poaching and illegal fishing, I introduce the green security game model which accounts for adversaries' behavior change and provide algorithms to plan effective sequential defender strategies. Second, I incorporate complex terrain information and design PAWS (Protection Assistant for Wildlife Security) which generates patrol routes to combat poaching. PAWS has been deployed in Southeast Asia for tiger conservation. In addition, I will cover our recent work on adversary behavior modeling and forecasting with real-world poaching data.
Bio: Fei Fang is a Postdoctoral Fellow at the Center for Research on Computation and Society (CRCS), Harvard University and an Adjunct Assistant Professor at the Institute for Software Research at Carnegie Mellon University. She received her Ph.D. from the Department of Computer Science at the University of Southern California in June 2016, advised by Professor Milind Tambe. She received her bachelor degree from the Department of Electronic Engineering, Tsinghua University in July 2011. Her research lies in the field of artificial intelligence and multi-agent systems, focusing on computational game theory with applications to security and sustainability domains. Her dissertation is selected as the runner-up for IFAAMAS-16 Victor Lesser Distinguished Dissertation Award. Her work has won the Innovative Application Award at Innovative Applications of Artificial Intelligence (IAAI'16), the Outstanding Paper Award in Computational Sustainability Track at the International Joint Conferences on Artificial Intelligence (IJCAI'15). Her work on "Protecting Moving Targets with Mobile Resources" has been deployed by the US Coast Guard for protecting the Staten Island Ferry in New York City since April 2013. Her work on designing patrol strategies to combat illegal poaching has lead to the deployment of PAWS application in a conservation area in Southeast Asia for protecting tigers.
Angela Fuller, U.S. Geological Survey, Cornell University
Fri May 19, 2017, 1:30-2:30pm EDT (UTC-4)
Title: Density-Weighted Connectivity for Landscape Management and Connectivity Conservation
Abstract: Many conservation efforts are focused on maintaining connectivity of protected areas or reserves as a biodiversity or species conservation strategy. The intended purpose of such corridors is to provide regions of the landscape that facilitate movement of individuals. Specific objectives include increasing gene flow, reducing isolation and inbreeding, increasing fitness and survival of species, and allowing species to move and adapt to changes in the landscape. Corridor conservation typically focuses on either 1) conserving areas that support high abundance of species to reduce the risk of demographic stochasticity or 2) conserving areas that allow individuals to move between reserve areas to maintain gene flow. Most corridor design applications focus on patterns of habitat and landscape structure (structural connectivity). However, the impetus of corridor design is the process of animal movement (functional connectivity). Functional connectivity considers the degree to which the landscape facilitates or impedes the movement of organisms and is the product of landscape structure and the response of organisms to this structure. However, maintenance of spatially structured populations requires considerations of both species abundance as well as functional landscape connectivity. I present a model for corridor design in the Chocó-Andean region of Ecuador, home to the endangered Andean bear (Tremarctos ornatus) and numerous endemic and threatened birds and describe a novel metric related to biodiversity conservation and corridor design. We use the ecological distance-based spatial capture-recapture model that simultaneously estimates species density and spatial aspects of animal population structure. The density-weighted connectivity metric is derived from encounter history data commonly collected in capture-recapture studies. I highlight how this metric can be used in reserve design or landscape management frameworks to inform conservation decision making.
Bio: Angela Fuller is the Leader of the New York Cooperative Fish and Wildlife Research Unit and an Associate Professor at Cornell University. Angela's research focuses on applied conservation and management of mammals, specifically related to population dynamics and the influence of human-induced landscape changes on populations. The second major program area of her research is applying structured decision making and adaptive management for aiding natural resource management and policy decisions. Her recent work has focused on informing agency decision making for managed species such as black bears, white-tailed deer, wild turkeys, and fishers; designing resilient and sustainable landscapes that support human quality of life and conserve biodiversity, with a focus on endangered Andean bears in Ecuador; and developing new methods for sampling and monitoring wildlife populations such as black bear, moose, mink, and fisher.
Doug Fisher, Vanderbilt University
Fri June 2, 2017, 1:30-2:30pm EDT (UTC-4)
Title: A survey of CompSustNet research, education, and outreach
Abstract: The talk will survey research, education, and outreach in CompSustNet. There will be unintentional gaps in my coverage of CompSustNet activities in this "draft talk", so there will be opportunities for the audience to help fill in the gaps in my coverage. There will also be opportunities to suggest how CompSustNet, and computational sustainability generally, might grow by filling in real gaps in its current coverage of research, education, and outreach. I will also survey other large centers and networks in related spaces of computing and/or sustainability, which might also inform CompSustNet's plans.
Bio: Doug Fisher is an Associate Professor of Computer Science at Vanderbilt University. His research has spanned unsupervised and supervised machine learning for prediction and problem solving, cognitive modeling, and more recently computational creativity. He served as a Program Director at NSF from 2007 - 2010, and was responsible for areas of AI (to include ML, MAS, KR, Planning) and he was a primary representative on sustainability for CISE. He received a Director's award in 2010 for all these activities. See his experience at NSF summarized at: http://www.cccblog.org/2011/08/24/first-person-life-as-a-nsf-program-director/. Doug has also worked in the online learning space, and was the founding Director of the Vanderbilt Institute for Digital Learning. He is the founding Faculty Director of Warren (residential) College at Vanderbilt University. Doug is the Director for Outreach, Education, Diversity, and Synthesis (OEDS) of CompSustNet.
Alan Fern, Oregon State University
Fri Jun 16, 2017, 1:30-2:30pm EDT (UTC-4)
Title: Bringing Bayesian Optimization into the Lab: Reasoning about Resources and Actions
Abstract: Bayesian optimization (BO) aims to optimize costly-to-evaluate functions by running a limited number of experiments that each evaluate the function at a selected input. Typical BO formulations assume that experiments are selected one at a time, or in fixed batches, and that experiments can be executed immediately upon request. This setup fails to capture many real-world domains where the execution of an experiment requires setup and preparation time. In this talk, we will present our research on extending the BO setting to incorporate these concerns. The result is a novel BO problem formulation that explicitly models the resources and actions needed to prepare and run experiments. Our algorithmic approach to this problem involves integrating BO principles with a Monte-Carlo tree search. A crucial ingredient is to exploit problem structure in order to design a heuristic function with approximation guarantees that can be used to effectively guide the search. Our experiments demonstrate the effectiveness of this approach and illustrate the more general promise of combining ideas from automated planning and BO.
Bio: Alan Fern is Professor of Computer Science and Associate Head of Research for the School of EECS at Oregon State University. He received his Ph.D. (2004) and M.S. (2000) in computer engineering from Purdue University, and his B.S. (1997) in electrical engineering from the University of Maine. He is an associate editor of the Machine Learning Journal, the Journal of Artificial Intelligence Research, and serves on the executive council of the International Conference on Automated Planning and Scheduling. His research interests span a range of topics in artificial intelligence, including machine learning and automated planning/control, with a particular interest in the intersection of those areas.
2017 Fall Seminar Details
Kevin Leyton-Brown, University of British Columbia
Fri Oct 20, 2017, 1:30-2:30pm EDT (UTC-4)
Title: Kudu: An Electronic Market for Agricultural Trade in Uganda
Abstract: Most adults in developing countries (e.g., 80% of Ugandans) are farmers. In such parts of the world, market inefficiency is a significant societal problem, arising substantially from poor access to information and other power differentials between farmers and buyers. Kudu is an electronic market for agricultural trade in developing countries designed to increase efficiency. Farmers and buyers enter bids using SMS and USSD, the messaging technologies available on non-internet-enabled phones. Kudu automatically identifies profitable trades and proposes them to participants by text message. As a positive side effect, Kudu infers accurate information about prevailing prices, which it broadcasts widely. For the past two years Kudu has been piloted in Uganda. Tens of thousands of users have registered with the system, and at least $2 million USD in crops have been traded. This talk will focus on the computational, incentive, and practical challenges we have faced designing a market in this unique setting.
Bio: Kevin Leyton-Brown is a professor of Computer Science at the University of British Columbia and an associate member of the Vancouver School of Economics. He holds a PhD and M.Sc. from Stanford University (2003; 2001) and a B.Sc. from McMaster University (1998). He studies the intersection of computer science and microeconomics, addressing computational problems in economic contexts and incentive issues in multiagent systems. He also applies machine learning to the automated design and analysis of algorithms for solving hard computational problems.
He has co-written two books, "Multiagent Systems" and "Essentials of Game Theory," and over 100 peer-refereed technical articles; his work has received over 8,000 citations and an h-index of 37. He is the recipient of UBC's 2015 Charles A. McDowell Award for Excellence in Research, a 2014 NSERC E.W.R. Steacie Memorial Fellowship—previously given to a computer scientist only 10 times since its establishment in 1965—and a 2013 Outstanding Young Computer Science Researcher Prize from the Canadian Association of Computer Science. He and his coauthors have received paper awards from JAIR, ACM-EC, AAMAS and LION, and numerous medals for the portfolio-based SAT solver SATzilla at international SAT solver competitions (2003–15).
He has co-taught two Coursera courses on "Game Theory" to over half a million students, and has received awards for his teaching at UBC—notably, a 2013/14 Killam Teaching Prize. He is chair of the ACM Special Interest Group on Electronic Commerce, which runs the annual Economics & Computation conference. He serves as an associate editor for the Artificial Intelligence Journal (AIJ), ACM Transactions on Economics and Computation (ACM-TEAC), and AI Access; serves as an advisory board member for the Journal of Artificial Intelligence Research (JAIR, after serving as associate editor for eight years); and was program chair for the ACM Conference on Electronic Commerce (ACM-EC) in 2012. In 2016 he was a visiting researcher at Microsoft Research New England and a visiting professor at Harvard's EconCS group. Previously, he spent the fall of 2015 at the Simons Institute for the Theory of Computing at UC Berkeley, and split his 2010–11 sabbatical between Makerere University in Kampala, Uganda and the Institute for Advanced Studies at Hebrew University of Jerusalem, Israel. He currently advises Auctionomics, Inc. (and through them, the Federal Communications Commission) and Qudos, Inc. He is a co-founder of Kudu.ug and Meta-Algorithmic Technologies. In the past, he served as a consultant for Zynga, Inc., Trading Dynamics Inc., Ariba Inc., Cariocas Inc., and was scientific advisor to UBC spinoff Zite Inc. until it was acquired by CNN in 2011.
Paul L. Fackler, North Carolina State University
Fri Nov 3, 2017, 1:30-2:30pm EDT (UTC-4)
Title: Solving stochastic dynamic programming models without transition matrices
Abstract: Discrete dynamic programming, widely used in addressing optimization over time, suffers from the so-called curse of dimensionality, the exponential increase in problem size as the number of system variables increases. One method to reduce the computational resources required to find solutions is to avoid the use of state transition probability matrices, which grow in the square of the size of the state space. This can be done through the use of expected value (EV) functions, which compute the expectation of functions of the future state variables conditioned on current variables. Two ways that this leads to potential gains arise when the state transition can be broken into separate phases and when the transitions for different state variables are conditionally independent. Both of these situations arise in models that are used in natural resource management and are illustrated with several examples including the dynamic reserves site selection problem, managing invasive species on a spatial network and managing wildlife harvests with multiple population stage classes. Efficiency gains include far lower memory requirements and orders of magnitude reductions in computing time.
Bio: Paul L. Fackler is a professor of agricultural and resource economics and associate professor of applied ecology at North Carolina State University and an internationally recognized teacher and scholar in the areas of decision analysis and computational methods. He co-authored a widely used textbook on the use of computational methods (Applied Computational Economics and Finance) along with the CompEcon Toolbox, a package of computer programs used in both teaching and research. The main focus of his research currently is the application of dynamic optimization tools to problems involving the management of natural resources. He is also the developer of the MDPSolve package for solving dynamic optimization problems.
Kristian Kersting, Technical University Darmstadt
Fri Nov 17, 2017, 1:30-2:30pm EST (UTC-5)
Title: Feeding the World with Big Data: Uncovering Spectral Characteristics and Dynamics of Stressed Plants
Abstract: Understanding the adaptation process of plants to (a)biotic stress is essential for improving management practices and breeding strategies of crops for sustainable agriculture in the coming decades. In this context, plant phenotyping is a main bottleneck in basic plant sciences and plant breeding, as it links genomics with complex responses of plants to varying environments. In particular, hyperspectral imaging is a promising approach for non-invasive, data-driven phenotyping, which allows for discovering non-destructively spectral characteristics of plants correlated with internal structure and physiological states in time-course experiments.
Unfortunately, data-driven phenotyping also presents unique computational problems in scale and interpretability: (1) Data is often gathered at massive scale, and (2) researchers and experts of complementary skills have to cooperate in order to develop models and tools for data intensive discovery that yield easy-to-interpret insights for users who are not necessarily trained computer scientists. On the problem of mining hyperspectral images to uncover spectral characteristic and dynamics of stressed plants, I will showcase that both challenges can be met and that big data mining can—and should—play a key role for feeding a hungry world, while enriching and transforming data mining.
Bio: Kristian Kersting is a Professor (W3) for Machine Learning at the CS Department of the TU Darmstadt, Germany, where he heads the machine learning lab. After receiving his Ph.D. from the University of Freiburg in 2006, he was with the MIT, Fraunhofer IAIS, the University of Bonn, and the TU Dortmund University. His main research interests are statistical relational AI, machine learning, and data mining, as well as their applications. Kristian has published over 150 peer-reviewed technical papers and co-authored a book on statistical relational AI. He received the European Association for Artificial Intelligence (EurAI, formerly ECCAI) Dissertation Award 2006 for the best AI dissertation in Europe as well as two best paper awards. He regularly serves on the PC (often at senior level) for several top conference and co-chaired ECML PKDD 2013 and UAI 2017. For more information please visit http://www.ml.informatik.tu-darmstadt.de/.
Michela Milano, University of Bologna
Fri Dec 1, 2017, 1:30-2:30pm EST (UTC-5)
Title: Empirical Model Learning for Sustainability Challenges
Abstract: Sustainability challenges often involve reasoning, managing and deciding on complex, interconnected global systems spanning across different sectors: examples include energy networks, financial markets, natural ecosystems, and cities.
In principle, using optimization methods to support high-level decision making activities in those contexts may lead to dramatically more sustainable and efficient policies. In practice, the kind of systems considered in computational sustainability can be an optimizer's worst nightmare: they typically involve complex infrastructures, organizations, laws and processes; they influence and are influenced by the environment, and are strongly perturbed by human behavior; they feature multiple actors, often self-interested and with conflicting objectives. The classical, expert-driven, modeling approach used in optimization has a very hard time coping with situations where the experts themselves are incapable of providing precise, non-ambiguous definitions of the problem constraints and goals.
We argue that dealing with systems of such a complexity calls for a strong integration of data science and optimization, raising interest in technique that try and bridge the gap between the two fields. Empirical Model Learning (EML) is one such technique: it is a methodology for learning model components directly from data and for actively using these components to prune the search space or guiding the solution process.
We basically have to learn relations between decidables (alternative decisions we can take) and a observables of interest. The data for the learning process can come from historical measurements or be collected by running simulations. These relations can be extracted in the form of classical Machine Learning models (e.g. Neural Networks, Decision Trees), and EML defines methods to cast such models into constraints and objective functions that can be readily incorporated into existing optimization technology.
The EML methodology is relatively recent and still an active research topic, but it has already been proved applicable to practical problems. In this talk, we will provide examples in the field of computational sustainability, along with empirical models learnt using different Machine Learning methods and integrated in different optimization techniques (Constraint Programming, Mixed Integer Non-Linear Programming, and SMT).
Bio: Michela Milano, PhD, is full professor of Intelligent Systems at the University of Bologna. Her research activity covers methodologies and techniques for the design and development of decision support systems in several areas, including sustainability problems, smart cities, policy making, industrial applications, and high performance computing. She is member of the executive committee of the Italian Association of Artificial Intelligence, board member of the European Association of Artificial Intelligence EurAI and Councilor of the Association for the Advancements of Artificial Intelligence AAAI. She has a broad research activity in the area of Artificial intelligence as demonstrated by over 160 publications in international journals and conferences. She is Editor in Chief of the Constraints Journal, Area Editor of INFORMS Journal on Computing and Constraint Programming Letters. Michela Milano has coordinated and participated in many national and EU-funded projects, like ePolicy in the field of sustainable policy making, COLOMBO for improving traffic in urban areas, DAREED for optimizing energy districts, OPRECOMP on transprecision computing. Michela Milano has been the recipient for a Google Faculty Award in 2016 for the integration of deep learning techniques in combinatorial models. She has many collaborations and research projects with SME and big industrial players. She is co-founder of the university spin-off MindIT.
Nicole Sintov, The Ohio State University
Fri Dec 15, 2017, 1:30-2:30pm EST (UTC-5)
Title: Theory + practice: Promoting sustainable behavior in real-world settings
Abstract: Collaborations between academics and practitioners can bring rich opportunities to concurrently advance theory and make practical impacts. However, numerous well-documented barriers hinder such collaborations, yielding limited examples of success and a need for lessons learned to guide future efforts. This talk will cover several collaborative studies in which academics and practitioners teamed up to successfully design and deliver interventions aimed at promoting sustainable behavior in residential settings. Findings advance theory and have practical implications. For instance, one study leveraged smart grid infrastructure as part of an energy reduction intervention. It explores whether a competition-based intervention fosters group cohesion in a group residential setting, as well as the influence of changes in group cohesion on energy conservation behaviors. Another study focuses on behavioral spillover, or the phenomenon observed when engaging in an initial target behavior, often the target of a behavioral intervention, is linked to performance of another, seemingly unrelated behavior. Using data from a longitudinal field experiment, this study investigates whether beginning to compost results in spillover to household waste prevention behaviors, including food, energy, and water waste prevention. In addition to covering intervention results, theoretical implications, and practical impacts, lessons learned for successful academia-practitioner relationships will be shared.
Bio: Dr. Nicole Sintov is Assistant Professor of Behavior, Decision-Making, and Sustainability in the School of Environment and Natural Resources. She is also an OSU Sustainable and Resilient Economies Discovery Theme affiliate. As an environmental psychologist, her work focuses on designing and evaluating interventions aimed at encouraging pro-environmental behavior. Additionally, she investigates psychosocial processes of change that occur during behavioral interventions to uncover truths about more than just which behaviors change, but also how and why such changes occur. She pursues these objectives as they pertain to a variety of domain areas, including building energy use, sustainable technology adoption, and the food-energy-water nexus. Her work has been supported by the National Science Foundation, US Department of Energy, Los Angeles Department of Water and Power, World Wildlife Fund, and US Army Research Office. She holds a B.S. in Psychology / Ecology from the University of California, San Diego, and a Master's in Psychology, graduate certificate in Sustainable Cities, and Ph.D. in Psychology from the University of Southern California. Learn more about the work in Nicole's research group here.