Dr. Sam Burer
1:00 - 2:00 p.m. CST
C125 PBB and via Zoom

Topic: Two Decades of Low-Rank Optimization

Abstract: Big data matrices are often low rank, and so modern optimization algorithms often require the use of low-rank decision variables. Such variables also arise naturally in certain optimization problems, irrespective of the input data. As a result, over the past two decades, low-rank optimization has grown as a critical area of research in optimization and machine learning. However, enforcing low-rank structure on a variable gives rise to severe nonconvexities, challenging the design of high-quality algorithms. In this talk, we offer our personal perspective on low-rank algorithms for semidefinite programs and related matrix optimizaton problems. We touch on their history and describe connections with other trends in optimization, including first-order methods and benign nonconvexity. Finally, we highlight recent theoretical insights into low-rank algorithms - as well as various applications for which low-rank approaches have proven successful.

Bio: Sam Burer is the Tippie Rollins Professor in the Department of Business Analytics at the University of Iowa. He received his Ph.D. from the Georgia Institute of Technology, and his research focuses on convex optimization, especially semidefinite and copositive programming. He is the recipient of the 2020 INFORMS Computing Paper Prize and the 2023 SIAM Optimization Test of Time Award. His work has been supported by grants from the National Science Foundation, including the CAREER award, and he currently serves as an area editor of Operations Research and as an associate editor for SIAM Journal on Optimization and Mathematical Programming. He also serves as Treasurer of the Mathematical Optimization Society and is a past Vice Chair of the SIAM Activity Group on Optimization.

VIEW THE VIDEO

Dr. Robert Nowak
1:00 - 2:00 p.m. CST
C125 PBB and via Zoom

Topic: Active Machine Learning: Combining Human and Artificial Intelligence for Improved Learning Efficiency and Accuracy

Abstract: Active machine learning combines the power of artificial intelligence with human intelligence to improve the efficiency and accuracy of machine learning algorithms. The fundamental idea behind active learning is to actively engage the human user in the learning process by selecting the most informative data points for labeling, thus reducing the amount of labeled data required to achieve a given level of accuracy. In this talk, we will discuss the theory and applications of active machine learning, including various sampling strategies, query selection methods, and active learning algorithms. We will also explore how active learning can be applied to different domains, such as crowdsourcing and recommendation systems, AI-assisted education, computer vision, and healthcare, and discuss the challenges and limitations of active learning techniques. Finally, we will highlight some recent developments in the field and provide some directions for future research. By attending this talk, participants will gain a better understanding of the principles of active machine learning and its potential applications in real-world scenarios.

Bio: Robert Nowak holds the Nosbusch Professorship in Electrical and Computer Engineering at the University of Wisconsin-Madison, where  he directs the AFOSR/AFRL University Center of Excellence on Data Efficient Machine Learning. His research focuses on machine learning, optimization, signal processing, and statistics. He serves on the editorial boards of the SIAM Journal on the Mathematics of Data Science and the IEEE Journal on Selected Areas in Information Theory.

VIEW THE VIDEO

Dr. Teresa Wu
1:00 - 2:00 p.m. CST
C125 PBB and via Zoom

Topic: Deep Learning to identify MRI signatures along the age continuum using Brain Age

Abstract: Imaging biomarkers are being increasingly applied for early diagnosis and staging of disease in humans. Developing imaging biomarkers requires advances in both image acquisition and analysis. In recent years, deep learning has rapidly dominated the computer vision field with advances also diffusing into the medical field. The objective of this study is to assess a deep learning approach to characterizing brain age signatures using MRI from cognitively normal subjects and explore its potential as biomarkers in neurodegenerative disease (e.g., Alzheimer’s disease) diagnosis. Two 3D deep ResNet-18 models were implemented to predict the chronological age of the healthy subjects. Both models were trained on 7372 T1 MRI from a combined lifespan cohort of 5848 cognitively normal participants (age: 8- 95 yrs). The first ResNet model was trained to regress the chronological age of participants on 3D MRI scans with a linear layer as the last layer. Different from that, we trained the other ResNet model for a multi-class classification where chronological age values were discretized into 86 classes and the last layer was a fully connected layer. The regression model achieved an MAE=3.76 years whereas the classification model achieved an MAE=2.65 years on same lifespan cohort. Both ResNet models were able to achieve state-of-art performance in predicting brain age. Using a mean-variance loss and translating the age prediction task into multi-class classification, the performance of brain age prediction was improved. We further derived the brain age signatures from ADNI (Alzheimer’s Disease Neuroimaging Initiative) cohort and observed group differences between Mild Cognitive Impairment (MCI) and Healthy Control (p<0.05). This supported our hypothesis that brain signatures have the potential to support neurodegenerative disease diagnosis.

Bio: Teresa Wu is a Professor from School of Computer and Augmented Intelligence (SCAI), Arizona State University and an adjunct Professor of Radiology in College of Medicine, Mayo Clinic. Her current research interests include imaging informatics and clinical decision support. Professor Wu has published ~140 journal articles in journals such as NeuroImage, NeuroImage: Clinical, Brain Communications, IEEE Transactions on Pattern Analysis and Machine Intelligence, and Information Sciences. Professor Wu is the founding Director of the ASU-Mayo Center for Innovative Imaging. She received numerous awards including NSF CAREER award (2003), AFOSR Summer Faculty Fellow (2010, 2011), IBM Faculty Award (2017), IISE Fellow (2020) and ASU PLuS Fellow on Global Health (2016-2020). She was a former Editor-in-Chief for IISE Transactions on Healthcare Systems Engineering (2016-2020).

VIEW THE VIDEO

Dr. Nitesh Chawla
1:00 - 2:00 p.m. CST
C125 PBB and via Zoom

Topic: Interdisciplinarity of Data Science and AI: Driving Innovation and Advancing Societal Impact

Abstract: The ubiquity of AI and data science is spurred by significant factors, including 1) major methodological and computational advances; 2) the increasing availability of problem statements in sciences, engineering, business and the social sciences; and 3) an increasing demand from academia, research institutions, corporations, and governments for talent. This  is presenting unparalleled opportunities to accelerate scientific discovery, innovation and translation. In this talk, I’ll present elements of our research program and vision that is centered on the principles of data & society, including partnerships, fostering innovation in data science and AI while generating educational opportunities, societal impact, and entrepreneurship.

Bio: Nitesh Chawla is the Frank M. Freimann Professor of Computer Science and Engineering and the Founding Director of the Lucy Family Institute for Data and Society at University of Notre Dame. His research is focused on artificial intelligence, data science, and network science, and is motivated by the question of how technology can advance the common good through interdisciplinary research.  His publications have received more than 50,000 citations with an h-index of 75. He is a Fellow of the Institute of Electrical and Electronics Engineers (IEEE), Fellow of the Association of Computing Machinery (ACM) and Fellow of the Asia Pacific Association for Artificial Intelligence (AAIA). He is the recipient of multiple awards including National Academy of Engineers New Faculty Fellowship, IEEE CIS Outstanding Early Career Award, Rodney F. Ganey Community Impact Award, IBM Big Data & Analytics Faculty Award, and the 1st Source Bank Technology Commercialization Award.  He is co-founder of Aunalytics, a data science software and cloud computing company.

VIEW THE VIDEO

Dr. Patrick Fan
11:00 a.m. CST
W401 PBB

Topic: Idea Recommendation in Open Innovation Platforms: A Design Science Approach

Abstract: Collaborative crowdsourcing communities help firms obtain ideas generated by the public at a lower cost compared to those generated in-house. However, the growth of these communities has led to a large influx of ideas of mixed quality, which has made it difficult for firm experts to select and implement ideas. In this paper we propose a novel theoretical framework to (1) extract important features from a collaborative crowdsourcing community and (2) apply them to the practice of recommending ideas that are most likely to be implemented in the future. More specifically, we adopt the design science research paradigm, introduce the knowledge persuasion model as the kernel theory, operate users’ persuasive attempts and firm experts’ persuasive coping, and identify a rich set of features as predictors of the likelihood of idea implementation. We test our prediction framework on a large-scale collaborative crowdsourcing community. The results of our data analysis show that the proposed framework is effective and efficient in predicting the likelihood of idea implementation. To increase the interpretability of the prediction model, we also implement the SHapley Additive exPlanations (SHAP) analysis and discuss the relationships between important features and idea implementation. We conclude by discussing the theoretical and practical implications of these findings.

Bio: Dr. Weiguo (Patrick) Fan is Henry B. Tippie Chair in Business Analytics at the Tippie College of Business, University of Iowa. He received his Ph.D. in Business Administration from the Ross School of Business, University of Michigan, Ann Arbor, in 2002, a M. Sce in Computer Science from the National University of Singapore in 1997, and a B. E. in Information and Control Engineering from the Xi'an Jiaotong University, P.R. China, in 1995.

His research interests focus on the design and development of novel information technologies - information retrieval, data mining, text/web mining, business intelligence techniques, data science, business analytics - to support better business information management and decision making. He has also worked on causal inference on observational data using econometric models and field experiments in e-Commerce settings. His research findings have been widely cited and used in academia and industry.

His research has been cited more than 12000 times (H-index: 53) according to Google Scholar. His research has been funded by five USA NSF grants, three China NSF grants, one PWC grant, one KPMG grant, and several other industrial grants. He currently serves on the editorial boards for several well-known IS journals: MIS Quarterly, Journal of Association for Information Systems, Information Systems Journal, Information and Management, Information Technology and Management, Journal of Electronic Commerce Research and Journal of Database Management.

He has consulted with many Fortune 500 companies. He also has served as reviewers and advisors for NSF, National Homeland Security, and many startups.

view the video

Dr. Rong Jin
Alibaba
11:00 a.m. CST

Topic: Future AI: Bridge the Gap between Machine Intelligence and Human Intelligence

Abstract: Although we have witnessed amazing progress in deep learning, there are many fundamental questions related to deep learning that remain to be answered. In particular, even though deep learning framework is inspired by how human brains process information, a significant gap needs to fill out between machine intelligence empowered by deep learning and human intelligence. In this talk, we first examine the main limitations of deep learning in comparison with human intelligence, including a large gap in sample complexity, limited abilities in exploring knowledge and logic for effective inference, and large difference in information processing and network structure. We then present recent progress in these dimensions that may shed lights on how to bridge the gap between human intelligence and machine intelligence.

Bio: Rong Jin is currently an associate director of DAMO academy at Alibaba, leading the research and development of state-of-the-art AI technologies. Before joining Alibaba, he was a faculty member of the Computer and Science Engineering Dept. at Michigan State University from 2003 to 2015. His research is focused on statistical machine learning and its application to large-scale data analysis. He published over 300 technique papers, mostly on the top conferences and prestigious journals. He is an associate editor of IEEE Transaction at Pattern Analysis and Machine Intelligence (TPAMI) and ACM Transaction at Knowledge Discovery from Data. Dr. Jin holds Ph.D. in Computer Science from Carnegie Mellon University. He received the NSF career award in 2006.

VIEW THE VIDEO

Dr. Ruslan Salakhutdinov
Carnegie Mellon University
11:00 a.m. CST

Topic: From Differentiable Reasoning to Self-supervised Embodied Active Learning

Abstract: In this talk, I will first discuss deep learning models that can find semantically meaningful representations of words, learn to read documents, and answer questions about their content. I will introduce methods that can augment neural representation of text with structured data from Knowledge Bases (KBs) for question answering, and show how we can answer compositional questions over long structured documents using a text corpus as a virtual KB. In the second part of the talk, I will show how we can design modular hierarchical reinforcement learning agents for visual navigation that can perform tasks specified by natural language instructions, perform efficient exploration and long-term planning, build and utilize 3D semantic maps to learn both action and perception models in self-supervised manner, while generalizing across domains and tasks.

Bio: Russ Salakhutdinov is a UPMC Professor of Computer Science in the Department of Machine Learning at CMU. He received his PhD in computer science from the University of Toronto. After spending two post-doctoral years at MIT, he joined the University of Toronto and later moved to CMU. Russ's primary interests lie in deep learning, machine learning, and large-scale optimization. He is an action editor of the Journal of Machine Learning Research, served as a program co-chair for ICML2019, and served on the senior programme committee of several top-tier learning conferences including NeurIPS and ICML. He is an Alfred P. Sloan Research Fellow, Microsoft Research Faculty Fellow, Canada Research Chair in Statistical Machine Learning, and a recipient of the Early Researcher Award, Google Faculty Award, and Nvidia's Pioneers of AI award.

Dr. Zhongju (John) Zhang
Arizona State University
11:00 a.m. CST
S401 PBB

Topic: What Types of Crowds Generate More Valuable Content? Evidence from Cross-Platform Posting

Abstract: This study examines the value of content generated by various groups of users on a financial social media platform. The value of such content is measured by the incremental accuracy of predicting stock volatility based on the cues from those content. We argue that the characteristics of a crowd such as crowd size, crowd diversity, and crowd independence have significant impact on the predictive value of the crowd-generated content. Leveraging a natural experiment setup where the financial platform no longer receives cross-postings from another major social media platform, we show empirical evidence that the composition of crowds (i.e., characteristics of the users who make up crowds) does matter. Furthermore, the impacts of the characteristic features on the value of crowd-generated content are likely heterogeneous and non-monotone. Finally, we discuss the implications of this study about digital platform envelopment and competition.

Bio: Zhongju Zhang is Professor of Information Systems and Data Analytics at the W. P. Carey School of Business, Arizona State University. Prior to that, Zhang was a tenured faculty member and the founding director of the MS degree program in business analytics and project management (MS-BAPM) at the School of Business, University of Connecticut. Zhang has formal training in the fields of information systems, data science, computer science, management science, operations research/management, and economics. He has extensive knowledge and strongly believes in experiential learning in higher education. Zhang has consulted extensively for various industries on data centric, operations, and strategic decision problems, and has experience developing strategic analytics initiatives for organizations. Zhang has also successfully sought external funding from both the industry and government agencies.

Zhang’s research primarily focuses on how information systems/technology and data analytics impact consumer behavior, create business value, and transform business models. He has specific interests in online collaborative platforms and social media, fraud and fake news detection as well as its business and social impacts, platform economy and business models, economic aspects of digital transformation, machine learning and data science. Zhang's work has appeared in academic journals including Information Systems Research, MIS Quarterly, Production and Operations Management, INFORMS Journal on Computing, Journal of Management Information Systems, Decision Support Systems, European Journal of Operational Research, and Decision Sciences. Zhang serves on the Senior Editorial Boards of a few journals including Production and Operations Management, Decision Support Systems, and Electronic Commerce Research and Applications. He has won numerous research and teaching awards, including the Best IS Publications of the Year Award, Research Excellence Award, MBA Teacher of the Year Award, and the Ackerman Scholar Award.

VIEW THE VIDEO

Dr. Amos Storkey
University of Edinburgh
11:00 a.m. CST

Topic: Towards a Distributed Transactional Artificial Intelligence

Abstract: The story of recent machine learning and artificial intelligence has been one of bigger and bigger models, and accumulation and use of more and more data. Improvements in current measures of performance though this approach are undeniable; some complex relationships cannot be discovered without more data to inform that discovery. However this emphasis on large data accumulation as the driver for artificial intelligence has led to natural monopolies and centralization of resource in the implementation of artificial intelligence. The story of machine learning in the 21st century has to a large extent been a story of accumulation and size. Yet in many fields waves of centralization are followed by waves of decentralization and distributed mechanisms -- whether it be market versus control economies or moves from mainframes to personal computing to cloud compute. I am persuaded the next piece in the story of artificial intelligence will be, and needs to be, a decentralised transactional artificial intelligence.

In this talk I discuss some of the technical pieces of the puzzle that will be key in the process of formulating a distributed artificial intelligence through autonomous agents, from meta-learning, multi-task and multi-agent learning through to economic and prediction-market models of machine learning, as well as discussing some of the unsolved problems in mechanism design for a distributed transactional artificial intelligence.

Bio: Amos Storkey is Professor of Machine Learning and AI in the School of Informatics, University of Edinburgh, with a background in mathematics (MA Mathematics, Trinity, Cambridge) and theoretical physics (Part III Mathematics) before focusing on Machine Learning (MSc, PhD Imperial London). He moved to Edinburgh after his PhD, where he now leads the Bayesian and Neural Systems Group focused on deep neural networks, transfer learning, probabilistic models, Bayesian methods, transactional machine learning and efficient learning and inference. He is known for the Storkey Learning Rule for Hopfield networks, for a first neural network capable of competitive Go play learned only from human play, for market models of machine learning and for his broader contributions to Bayesian methods, meta-learning, efficiency in neural networks and curiosity driven reinforcement learning. He is currently director of the UK EPSRC Centre for Doctoral Training in Data Science.

VIEW THE VIDEO

Dr. Asuman Ozdaglar
MIT
11:00 a.m. CST

Topic: Optimal and Differentially Private Data Acquisition from Strategic Users

Abstract: The data of billions of people around the world are used every day for improving search algorithms, recommendations on online platforms, personalized advertising, and the design of new drugs, services, and products. With rapid advances in machine learning (ML) algorithms and further growth in data collection, these practices will become only more widespread in the years to come. A common concern with many of these data-intensive applications centers on privacy — as a user’s data is harnessed, more and more information about her behavior and preferences are uncovered and potentially utilized by platforms and advertisers. A popular solution to the tension between privacy costs and benefits of data is to use methods such as differential privacy in order to limit the extent to which an individual’s data is uncovered and exploited. Although these methods are already used by many of the tech companies, a key practical question remains: How do we decide how much privacy heterogeneous, strategic users will obtain?

This talk is an attempt to answer this key question and study the impact of data market architecture on the design of mechanisms for purchasing data from privacy sensitive strategic users. We consider a platform interested in estimating an underlying parameter using data collected from users. We formulate this question as a Bayesian-optimal mechanism design problem, in which an individual can share her (verifiable) data in exchange for a monetary reward or services, but at the same time has a (private) heterogeneous privacy sensitivity that represents her cost per unit privacy loss. We consider two popular differential privacy settings for providing privacy guarantees for the users: central and local. In both cases, we establish minimax lower bounds for the estimation error and derive (near) optimal estimators for given heterogeneous privacy loss levels for users. Building on this characterization, we pose the mechanism design problem as the optimal selection of an estimator and payments that will elicit truthful reporting of privacy sensitivities of users under a regularity condition on the distribution of privacy sensitivities. Our mechanism in the central setting can be implemented in log-linear time in the number of users, and, in the local setting, it admits a Polynomial Time Approximation Scheme (PTAS).

Bio: Asu Ozdaglar received the B.S. degree in electrical engineering from the Middle East Technical University, Ankara, Turkey, in 1996, and the S.M. and Ph.D. degrees in electrical engineering and computer science from the Massachusetts Institute of Technology, Cambridge, in 1998 and 2003, respectively.

She is the Mathworks Professor of Electrical Engineering and Computer Science (EECS) at the Massachusetts Institute of Technology (MIT). She is also the department head of EECS and deputy dean of academics of the Schwarzman College of Computing at MIT. Her research expertise includes optimization theory and algorithms, game theory, machine learning and network analysis with applications in social, economic and financial networks. Her recent research focuses on designing incentives and algorithms for data-driven online systems with many diverse human-machine participants. 

Professor Ozdaglar is the recipient of a Microsoft fellowship, the MIT Graduate Student Council Teaching award, the NSF Career award, the 2008 Donald P. Eckman award of the American Automatic Control Council, the 2014 Spira teaching award, and Keithley, Distinguished School of Engineering and Mathworks professorships. She is an IEEE fellow and was selected as an invited speaker at the International Congress of Mathematicians. She served on the Board of Governors of the Control System Society in 2010 and was an associate editor for IEEE Transactions on Automatic Control. She was the inaugural area co-editor for the area entitled "Games, Information and Networks” in the journal Operations Research. She is the co-author of the book entitled “Convex Analysis and Optimization” (Athena Scientific, 2003).

VIEW THE VIDEO

Dr. Gedas Adomavicius
University of Minnesota
11:00 a.m. CST

Topic: Recommender Systems and Preference Pollution

Abstract: Interactions between individuals and recommender systems can be viewed as a continuous feedback loop, consisting of pre-consumption and post-consumption phases. Pre-consumption, systems provide recommendations that are typically based on predictions of user preferences. They represent a valuable service for both providers and users as decision aids. After item consumption, the users often provide post-consumption feedback (e.g., preference ratings) to the system, which typically serves as the “ground truth” data that is used to further improve the system’s subsequent recommendations, completing the feedback loop. There is a growing understanding that this feedback loop can be a significant source of biases and unintended consequences. In particular, this talk focuses on “preference pollution”, a form of bias which reflects an unintended effect of system recommendations on the users’ post-consumption preference ratings. We provide a comprehensive exploration of the “preference pollution” phenomenon and discuss its importance and implications for the design and application of recommendation systems.

Bio: Gedas Adomavicius is a professor in the Department of Information and Decision Sciences at the Carlson School of Management, University of Minnesota, where he also holds the Larson Endowed Chair for Excellence in Business Education. He received his PhD degree in computer science from New York University. His general research interests revolve around computational techniques for aiding decision-making in information-intensive environments and include personalization technologies and recommender systems, machine learning and data analytics, and electronic market mechanisms. His research has been published in a number of leading academic journals in information systems and computer science, including Information Systems Research, MIS Quarterly, Management Science, Journal of Operations Management, IEEE Transactions on Knowledge and Data Engineering, ACM Transactions on Information Systems, and Data Mining and Knowledge Discovery, and has been cited more than 26,000 times to date (according to Google Scholar). He has received several research grants from major funding institutions, including the U.S. National Science Foundation CAREER award for his research on personalization technologies. He has served on the editorial boards of several leading academic journals, including as Senior Editor for Information Systems Research and MIS Quarterly. In 2017, Prof. Adomavicius received the INFORMS Information Systems Society’s Distinguished Fellow Award. At the Carlson School of Management, he has taught analytics-related courses in the undergraduate, MBA, MSBA, PhD, and Executive Education programs and is currently serving as the chair of the Information and Decision Sciences Department.

VIEW THE VIDEO

Dr. Xi Chen
New York University

Topic: Robust Online Learning and Its Applications to Assortment Optimization

Abstract: In this talk, we will first provide an overview of my research on online learning and decision-making. We will highlight a few applications on dynamic pricing, assortment optimization for online recommendation, and crowdsourcing.

Most online learning problems are built on an underlying probabilistic model. However, these models are inherently misspecified to a certain degree, which calls for robust learning methods. The second part of the talk will discuss robust online learning and its applications to recommendations. In particular, we consider the dynamic assortment optimization problem under the multinomial logit model (MNL) with unknown utility parameters. Based on online eps-contamination modeling of customers’ purchase behavior, we develop a rate-optimal robust online assortment optimization policy via an active elimination strategy.

Bio: Xi Chen is an associate professor at Stern School of Business at New York University, who is also an affiliated professor to Computer Science and Center for Data Science. Before that, he was a Postdoc in the group of Prof. Michael Jordan at UC Berkeley. He obtained his PhD from the Machine Learning Department at Carnegie Mellon University. He studies high-dimensional statistical learning, online learning, large-scale stochastic optimization, and applications to revenue management and crowdsourcing. He received an NSF Career Award and Outstanding Faculty Research Awards from Google, Facebook, Adobe, Alibaba, Bloomberg, and JPMorgan, and was featured in Forbes list of "30 Under 30 in Science."

VIEW THE VIDEO

Dr. Rong Chen
Rutgers University

Topic: New Approaches to Statistical Learning of Modern Time Series

Abstract: In the BIGDATA era, many new forms of data have become available and useful in various important applications. When these data are observed over time, they form new types of time series that require new statistical models and analytical tools in order to extract useful information. In this talk we present new developments in analyzing matrix/tensor time series, dynamic networks, functional time series and compositional time series, with applications ranging from economics, finance, international trade, electricity loading and others. We will also briefly discuss approaches on modeling other forms of time series, including text time series and dynamic social networks.

Bio: Dr. Chen is a Distinguished Professor of Statistics and Chair of the Department of Statistics at Rutgers University. His teaching and research interests include analysis of complex time series and dynamic systems, Monte Carlo methods and statistical applications in bioinformatics, business and economics, and engineering. He is an elected Fellow of the American Statistical Association and the Institute of Mathematical Statistics. Dr. Chen served as a co-editor of Journal of Business and Economic Statistics and is currently serving as co-editor of Statistica Sinica. He is former Treasurer of the Institute of Mathematical Statistics and former program director in the Division of Mathematical Sciences at the National Science Foundation. Dr. Chen received both his PhD and MS in Statistics from Carnegie Mellon University and his BS in mathematics from Peking University in China.

Dr. Melanie Mitchell
Santa Fe Institute

Topic: Abstraction and Analogy in Natural and Artificial Intelligence

Abstract: In 1955, John McCarthy and colleagues proposed an AI summer research project with the following aim: “An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves.” More than six decades later, all of these research topics remain open and actively investigated in the AI community. While AI has made dramatic progress over the last decade in areas such as vision, natural language processing, and robotics, current AI systems still almost entirely lack the ability to form humanlike concepts and abstractions.

In this talk, I will argue that the inability to form conceptual abstractions—and to make abstraction-driven analogies—is a primary source of brittleness in state-of-the-art AI systems, which often struggle in adapting what they have learned to situations outside their training regimes. I will reflect on the role played by analogy-making at all levels of intelligence, and on the prospects for developing AI systems with humanlike abilities for abstraction and analogy.

Bio: Melanie Mitchell is the Davis Professor at the Santa Fe Institute. Her current research focuses on conceptual abstraction, analogy-making, and visual recognition in artificial intelligence systems. Dr. Mitchell is the author or editor of six books and numerous scholarly papers in the fields of artificial intelligence, cognitive science, and complex systems. Her book Complexity: A Guided Tour (Oxford University Press) won the 2010 Phi Beta Kappa Science Book Award and was named by Amazon.com as one of the ten best science books of 2009. Her latest book is Artificial Intelligence: A Guide for Thinking Humans (Farrar, Straus, and Giroux).

VIEW THE VIDEO

Dr. Özlem Ergun
Northeastern University

Topic: Optimizing Post-Disruption Response Operations to Improve Resilience of Critical Infrastructure Systems

Abstract: Critical infrastructure systems (CIS) underpin almost every aspect of the modern society by enabling the essential functions through overlaying service networks. After a disruption impacting the CIS, the functionality of the overlaying service networks degrades. Thus, after an extreme event, in order to minimize the negative impact to society, it is crucial to restore the disrupted CIS as soon as possible. In this talk, we focus on disruptions created by natural hazards on transportation CIS and develop methods to efficiently plan the post-disaster response operations.

In the aftermath of a natural disaster, the transportation network is disrupted due to the debris blocking the roads and obstructing the flow of relief aid and search-and-rescue teams between critical facilities and disaster sites. In the first few days following a disaster, in order to deliver aid to those in need, blocked roads must be cleared by pushing the debris to the sides. In this context, we define the road network recovery problem (RNRP) as finding a schedule to clear the roads with limited resources such that all the service demanding locations are served in the shortest possible time. First, we address the deterministic RNRP and propose a novel network science inspired measure to quantify the criticality of the components within a disrupted service network and develop a restoration heuristic. Next, we consider RNRP with stochastic demand and propose an approximate dynamic programming approach for identifying an effective policy under uncertainty.

Bio: Dr. Özlem Ergun is a professor and Associate Chair for Graduate Studies in Mechanical and Industrial Engineering at Northeastern University. Dr. Ergun’s research focuses on design and management of large-scale and decentralized networks. She has applied her work on network design, management, and resilience to problems arising in many critical systems including transportation, pharmaceuticals, and healthcare. She has worked with organizations that respond to emergencies and humanitarian crises around the world, including USAID, UNWFP, UNHCR, IFRC, OXFAM America, CARE USA, FEMA, USACE, CDC, AFCEMA, and MedShare International. Dr. Ergun served as a member of the National Academies Committee on Building Adaptable and Resilient Supply Chains after Hurricanes Harvey, Irma, and Maria and is currently serving in the Committee on Security of America's Medical Product Supply Chain. Within INFORMS, Dr. Ergun has been a leader in establishing a strong community of OR/MS professionals with an interest in public programs. She was the President of INFORMS Section on Public Programs, Service and Needs in 2013. She currently serves as the Area Editor at the Operations Research journal for Policy Modeling and the Public Sector Area and the Department Editor at MSOM journal for Environment, Health and Society Department. Dr. Ergun is also a founding co-chair of the Health and Humanitarian Logistics Conference, held annually since 2009. In addition, Dr. Ergun was the Vice President of Membership and Professional Recognition on the INFORMS Board of Directors, 2011-2015. Prior to joining Northeastern she was the Coca-Cola Associate Professor in the School of Industrial and Systems Engineering at Georgia Institute of Technology, where she also co-founded and co-directed the Health and Humanitarian Systems Research Center at the Supply Chain and Logistics Institute. She received a BS in Operations Research and Industrial Engineering from Cornell University in 1996 and a PhD in Operations Research from the Massachusetts Institute of Technology in 2001.

VIEW THE VIDEO

Dr. Galit Shmueli
National Tsing Hua University

Topic: "Improving" Prediction of Human Behavior Using Behavior Modification

Abstract: The fields of machine learning and statistics have invested great efforts into designing algorithms, models, and approaches that better predict future observations. Larger and richer data have also been shown to improve predictive power. This is especially true in the world of human behavioral big data, as is evident from recent advances in behavioral prediction technology. Large internet platforms that collect behavioral big data predict user behavior for their internal commercial purposes as well as for third parties, such as advertisers, insurers, security forces, and political consulting firms, who utilize the predictions for user-level personalization, targeting, and other decision-making. While machine learning algorithmic and data efforts are directed at improving predicted values, the internet platforms can minimize prediction error by "pushing" users' actions towards their predicted values using behavior modification techniques. The better the platform is able to make users conform to their predicted outcomes, the more it can boast both its predictive accuracy and its ability to induce behavior change. Hence, internet platforms have a strong incentive to "make the prediction true", that is, demonstrate small prediction error. This strategy is absent from the machine learning and statistics literature. Investigating the properties of this strategy requires incorporating causal terminology and notation into the correlation-based predictive environment. However, such an integration is currently lacking. To tackle this void, we integrate Pearl's causal do(.) operator to represent and integrate intentional behavior modification into the correlation-based predictive framework. We then derive the Expected Prediction Error given behavior modification, and identify the components impacting predictive power. Our formulation and derivation make transparent the impact and implications of such behavior modification to data scientists, internet platforms and their clients, and importantly, to the humans whose behavior is manipulated. Behavior modification can make users' behavior not only more predictable but also more homogeneous; yet this apparent predictability is not guaranteed to generalize when the predictions are used by platform clients outside of the platform environment. Outcomes pushed towards their predictions can also be at odds with the client's intention, and harmful to the manipulated users.

Bio: Galit Shmueli is Tsing Hua Distinguished Professor at the Institute of Service Science, and Institute Director at the College of Technology Management, National Tsing Hua University, Taiwan. Before joining NTHU, she was the SRITNE Chaired Professor of Data Analytics and Associate Professor of Statistics & Information Systems at the Indian School of Business, and tenured Associate Professor at University of Maryland's Smith School of Business. She is the inaugural editor-in-chief of the INFORMS Journal on Data Science.

VIEW THE VIDEO

Dr. Andrea Lodi
Polytechnique Montréal

Topic: On the Interplay Between Machine Learning and Discrete Optimization

Abstract: In this talk, I will present my personal experience on the interplay between Machine Learning (ML) and Discrete Optimization by discussing three fruitful interactions. Moving from methodological research to applications, the first use case concerns ML-based branching for Mixed-Integer Linear Programming solvers, the second addresses stochastic prediction for tactical planning (applied to railways operations) and the third involves, at the real applied level, demand prediction and resource allocation during Covid.

Bio: As Canada Excellence Research Chair in Data Science for Real-Time Decision-Making at Polytechnique Montréal, Dr. Andrea Lodi holds Canada’s main chair in operations research. Dr. Lodi received the PhD in System Engineering from the University of Bologna in 2000 and he has been Herman Goldstine Fellow at the IBM Mathematical Sciences Department, NY in 2005–2006. He is author of more than 80 publications in the top journals of the field of Mathematical Optimization. He serves as Associate Editor for several prestigious journals in the area.

VIEW THE VIDEO