AAAI-17 Tutorial Forum
Thirtieth Conference on Artificial Intelligence
February 4–5, 2017, San Francisco, California, USA
What Is the Tutorial Forum?
The Tutorial Forum provides an opportunity for researchers and practitioners to spend two days each year freely exploring exciting advances in disciplines outside their normal focus. We believe this type of forum is essential for the cross fertilization, cohesiveness, and vitality of the AI field. We all have a lot to learn from each other; the Tutorial Forum promotes the continuing education of each member of AAAI.
(All tutorials are 4 hours, including breaks, unless otherwise noted.)
Saturday, February 4
9:00 AM – 1:00 PM
(except where noted; times include breaks, if applicable)
- SA1: Learn to Write a Scientific Paper of the Future: Reproducible Research, Open Science, and Digital Scholarship — Yolanda Gil, Daniel Garijo, Gail Peretsman-Clement
- SA2: Risk-Averse Decision Making and Control — Marek Petrik, Mohammad Ghavamzadeh
- SA3: Rulelog: Deep KRR for Cognitive Computing — Benjamin Grosof, Michael Kifer, Paul Fodor
- SA4: IoT Big Data Stream Minin — Gianmarco De Francisci Morales, Albert Bifet, Latifur Khan, Joao Gama, Wei Fan
- SA5: Computer Poker — Sam Ganzfried, Johannes Heinrich, Kevin Waugh
2:00 PM – 6:00 PM
(except where noted; times include breaks, if applicable)
- SP1: Recent Advances in Distributed Machine Learning — Wei Chen, Taifeng Wang, Tie-Yan Liu
- SP2: Statistical Relational Artificial Intelligence: Logic, Probability and Computation — Luc De Raedt, David Poole, Kristian Kersting, Sriraam Natarajan
- SP3: AI Planning for Robotics — Michael Cashmore, Daniele Magazzeni
- SP4: Modeling and Solving AI Problems in Picat — Roman Barták, Neng-Fa Zhou
2:00 PM – 3:45 PM
- SP5: AI for Data-Driven Decisions in Water Management — Biplav Srivastava, Sandeep S. Sandha
4:15 PM – 6:00 PM
- SP6: Social Data Bias in Machine Learning: Impact, Evaluation, and Correction — Huan Liu, Fred Morstatter
Sunday, February 5
9:00 AM – 1:00 PM
(except where noted; times include breaks, if applicable)
- SUA1: Deep Learning Implementations and Frameworks — Seiya Tokui, Kenta Oono, Atsunori Kanemura
- SUA2: Learning Bayesian Networks for Complex Relational Data — Oliver Schulte, Ted Kirkpatrick
- SUA3: Causal Inference and the Data-Fusion Problem — Elias Barenboim
- SUA4: Eliciting High-Quality Information — Boi Faltings, Goran Radanovic
- SUA5: Discrete Sampling and Integration for the AI Practitioner — Supratik Chakraborty, Kuldeep S. Meel, Moshe Y. Vardi
2:00 PM – 6:00 PM
(except where noted; times include breaks, if applicable)
- SUP1: Interactive Machine Learning: From Classifiers to Robotics — Matthew E. Taylor, Bradley H. Hayes, Ece Kamar
- SUP2: Knowledge Graph Construction from Text — Jay Pujara, Sameer Singh, Bhavana Dalvi
- SUP3: Introduction to multiAgent Path Finding — Glenn Wagner, Ariel Felner, Sven Koenig
- SUP4: Predicting Human Decision-Making: Tools of the Trade — Ariel Rosenfeld, Sarit Kraus
2:00 PM – 3:45 PM
- SUP5: Neuroevolution Reinforcement Learning — Risto Miikkulainen
4:15 PM – 6:00 PM
- SUP6: Artificial Intelligence and Video Games — Julian Togelius
SA1: Learn to Write a Scientific Paper of the Future: Reproducible Research, Open Science, and Digital Scholarship
Daniel Garijo, Yolanda Gil, and Gail Peretsman-Clement
This tutorial covers best practices in reproducible research, open science, and digital scholarship that help researchers increase citations for their papers, get credit for all their research products, augment their vitae with data and software that they have written, write compelling data management plans for funding proposals, comply with new funder and journal requirements, and practice open and reproducible science. We begin with a motivation for authors through an overview of why scientists, publishers, funders, and the public care about science practices. Next, we describe how to make research data accessible through publication in a public repository, including metadata, a license for reuse, and citable using a unique and persistent identifier. We then show how to make software accessible by making it available in a public repository, with a license, and a unique and citable persistent identifier. We also cover how to document in a software registry key information that helps others reuse research software. We then discuss how to document provenance and methods by explicitly describing related computations and outcomes in a workflow sketch, a formal workflow, or a provenance record, possibly with a persistent identifier. Finally, we provide a summary checklist for authors, and show how to manage their scholarly identity, reputation, and impact throughout their careers.
Tutorial materials are available on the supplemental tutorial site.
Daniel Garijo is a postdoctoral researcher at the Information Sciences Institute of the University of Southern California. His research activities focus on e-Science and the Semantic Web, specifically on how to increase the understandability of scientific experiments by publishing their outputs, inputs, provenance, metadata and intermediate results on the Web.
Yolanda Gil is director of knowledge technologies at the Information Sciences Institute, and a research professor in computer science at the University of Southern California. Her recent research focuses on knowledge capture in science, including semantic workflows, metadata, and provenance. She received her PhD in computer science from Carnegie Mellon University.
Gail Peretsman-Clement is a research library administrator with extensive experience in scientific, academic, and grey literature publishing; copyright education; authorship ethics; and E-research support. As head of research services at the California Institute of Technology Library, she leads a team responsible for research services, acquisitions, licensing, collection strategies, digital repository development, instruction and outreach, and research data services.
SA2: Risk-Averse Decision Making and Control
Marek Petrik and Mohammad Ghavamzadeh
Practical deployments often require risk-averse solutions, but most decision-making methods maximize the expected rewards. The operations research community has made significant progress in risk-averse decision-making by building on the conditional value at risk (CVaR) and the framework of convex risk measures. These principled approaches to risk-aversion are now also gaining traction in artificial intelligence and machine learning. Unfortunately, many recent advances on convex risk measures are not readily accessible to the artificial intelligence community because they are published in unfamiliar venues and are often presented in an overly technical manner. This tutorial will combine recent results from relevant communities to describe risk-sensitive methods in a way that is accessible to people with artificial intelligence and machine learning background. We will introduce and motivate convex measures of risk and describe efficient methods to optimize them in decision-making, control, planning, and reinforcement learning. We will also describe connections to robust optimization and to expected utility models, such as the relationship between entropic risk measures and exponential utility. The tutorial will include practical problems in which risk-sensitivity was used in both machine learning and operations research including the successes and difficulties with risk aversion in dynamic decision problems.
Marek Petrik is an assistant professor of computer science at the University of New Hampshire. He received his PhD from the University of Massachusetts Amherst in 2010. From 2011 to 2016 he was a research staff member at IBM's T.J. Watson Research Center.
Mohammad Ghavamzadeh is a senior analytics researcher at Adobe. He received a PhD in computer science from University of Massachusetts Amherst in 2005. From 2005 to 2008, he was a postdoctoral fellow at University of Alberta. He has been a permanent researcher at INRIA in France since November 2008.
SA3: Rulelog: Deep KRR for Cognitive Computing
Benjamin Grosof, Michael Kifer, and Paul Fodor
In this half-day tutorial, we cover the fundamental concepts, key technologies, emerging applications, recent progress, and outstanding research issues in the area of Rulelog, a leading approach to rule-based knowledge representation and reasoning (KRR). Rulelog matches well many of the requirements of cognitive computing. It combines deep logical/probabilistic reasoning tightly with natural language processing, and complements machine learning. Rulelog interoperates and composes well with graph databases, relational databases, spreadsheets, XML, and expressively simpler rule/ontology systems — and can orchestrate overall hybrid KRR. Developed mainly since 2005, Rulelog is much more representationally powerful than the previous state-of-the-art practical KRR approaches, yet is computationally affordable. It is fully semantic and has capable efficient implementations that leverage methods from logic programming and databases, including dependency-aware smart caching and a dynamic compilation stack architecture.
Rulelog extends Datalog (database logic) with general classical-logiclike formulas — including existentials and disjunctions — and strong capabilities for meta knowledge and reasoning, including higher-order syntax, flexible defeasibility and probabilistic uncertainty, and restraint bounded rationality that ensures worst-case polynomial time for query answering. A large subset of Rulelog is in draft as an industry standard. An exciting research frontier is that Rulelog can combine closely with machine learning and with natural language processing to both interpret and generate English, including potentially for conversational NL interaction.
The most complete system today for Rulelog is Ergo from Coherent Knowledge. A subset of Rulelog is also implemented in an open-source Flora-2 system and an earlier SILK system from Vulcan. Using Ergo, we will illustrate Rulelog's applications in deep reasoning and representing complex knowledge — such as policies, regulations/contracts, science, and terminology mappings — across a wide range of tasks and domains in business, government, and academe. Examples include: legal/policy compliance, for example, in financial services; financial reporting/accounting; health care treatment guidance and insurance; education/tutoring; security/confidentiality policies; and e-commerce marketing.
Prerequisite knowledge: Background assumed of participants is only the basics of first-order-logic and relational databases.
The target audience is those who are interested in learning more about knowledge representation and reasoning (KRR) as a core area in AI, and about how it relates to cognitive computing's applications as well as conceptual architecture. Together with machine learning (ML) and natural language interaction (NLI), KRR forms the tripod basis for the core of AI and cognitive computing. The audience will walk away with an understanding of Rulelog's key innovative logical and inferencing concepts, its broad applicability, its overall advantages and limitations, a sample of some specific application areas, and its open research topics.
Benjamin Grosof (lead presenter) principal director and research fellow in AI at Accenture, is a technical executive for an ambitious initiative on leveraging AI for business process automation across Accenture's $7.5B+ Operations group. He is an industry leader in AI knowledge representation, reasoning, and acquisition. He has pioneered semantic technology and industry standards for rules combined with ontologies, their acquisition from natural language, and their applications in finance, e-commerce, policies (including contracts, regulations, and security), and e-learning. He cofounded Coherent Knowledge, a software-centric startup that is commercializing a major research breakthrough in AI logical / probabilistic knowledge representation and reasoning (Rulelog) combined with natural language processing. Previously he was CTO and CEO of Coherent Knowledge (2013-2017; he is still a board member), and a senior research program manager at Vulcan Inc. (2007-2013), where he conceived and led the advanced research third of the predecessor of the Allen Institute for AI. Before Vulcan, he was an IT professor at MIT Sloan (2000-2007), where he was also a DARPA PI, and a senior software scientist at IBM Research (1988-2000). He cofounded the influential RuleML industry standards design effort and prototyped it in SweetRules, the main bases for the W3C Rule Interchange Format (RIF) standard. He cofounded the International Conference on Rules and Rule Markup Languages for the semantic web (which since became the RR and RuleML conferences and then the International Joint Conference on Rules and Reasoning). He led the invention of several fundamental technical advances in knowledge representation, including courteous defeasibility (exception-case rules), restraint bounded rationality (scalability in complex reasoning), and rule-based description logic ontologies (the basis for W3C's OWL RL standard). He also has extensive experience in user interaction design, and in combining logical methods with machine learning and probabilistic reasoning uncertainty. His background includes 5 major industry software releases, 5 years in software startups, his own part-time expert consulting firm, a Stanford PhD in AI, a Harvard BA in applied math, 3 patents, and over 60 refereed publications.
Grosof has given numerous invited talks about knowledge representation, reasoning, and acquisition, including with semantic rules, and developed several MIT courses with substantial focus on those. He presented with coauthors (including usually Michael Kifer since 2009) related tutorials on reasoning with complex knowledge at the AAAI Conference on Artificial Intelligence (2013), International Joint Conference on Artificial Intelligence (2001, 2016), the International Conference on Knowledge Capture (K-CAP 2015), International Semantic Web Conference (2004, 2005, 2006, 2009, 2010, 2012), the WWW conference (2006, 2009), the International Web Rule Symposium joint with the Reasoning Web Summer School (2015), and the ACM Conference on E-Commerce (2004).
Michael Kifer is a professor with the Department of Computer Science, Stony Brook University, USA. He received his PhD in computer science in 1984 from the Hebrew University of Jerusalem, Israel, and the MS degree in Mathematics in 1976 from Moscow State University, Russia.
Kifer is a cofounder of Coherent Knowledge, a new startup on semantic technology, and since 2012 he has been serving as the President of the Rules and Reasoning Association (RRA). His interests include Web information systems, knowledge representation, and database systems. He has published four text books and numerous articles in these areas. In particular, he coinvented F-logic, HiLog, and Transaction Logic, which are among the most widely cited works in computer science and, especially, in semantic web research. Kifer serves on the editorial boards of several computer science journals and chaired a number of conferences. Twice, in 1999 and 2002, he was a recipient of the prestigious ACM-SIGMOD "Test of Time" awards for his works on F-logic and object-oriented database languages. In 2013, Kifer's paper on transaction logic programming was awarded the Association of Logic Programming's Test of Time award as the most influential paper of 20 years ago. In 2006, Kifer was Plumer Fellow at Oxford University's St. Anne's College and, in 2008, he received SUNY Chancellor's award for excellence in scholarship. He has taught numerous courses at Stony Brook University since 1984.
Paul Fodor is a research assistant professor with the Department of Computer Science, Stony Brook University, USA. He received his PhD in computer science in 2011 from the Stony Brook University, New York, preceded by his MS degree in 2006 from Stony Brook University, and B.Sc. in computer science in 2002 from the Technical University of Cluj-Napoca, Romania.
Fodor is a cofounder of Coherent Knowledge with over 10 years' experience in databases research, natural language processing, artificial intelligence and stream processing systems. His work on declarative rule languages and logic used as a specification language and implementation framework for knowledge bases was applied in areas ranging from natural language processing to complex event processing and semantic web technologies. Through his research, DFodor has contributed to several large software projects: the IBM Watson natural language processing system for the Jeopardy! Challenge with human champions, the OpenRuleBench suite of benchmarks for analyzing the performance and scalability of rule systems for the semantic Web, the ETALIS declarative complex event processing and stream reasoning system, and the SILK Semantic Inferencing on Large Knowledge. Fodor was principal investigator, co-PI and contractor for projects funded by both public governmental sources and private companies such as PI for the SILK project funded by Vulcan Inc. to develop intelligent textbooks, contractor for the IBM Watson project, contractor for XSB Inc. for the DARPA Component, context, and manufacturing model library (C2M2L-1) using XSB Prolog, and PI for the Stony Brook University Hospital's Lung Cancer Evaluation Center management program. He has taught numerous courses at Stony Brook University since 2011.
SA4: IoT Big Data Stream Mining
Gianmarco De Francisci Morales, Albert Bifet, Latifur Khan, Joao Gama, and Wei Fan
Internet of things (IoT) has been recognized as one of the most exciting and key opportunities for both academia and industry. Advanced analysis of big data streams from sensors and devices is bound to become a key area of data mining research as the number of applications requiring such processing increases. Dealing with the evolution over time of such data streams, that is, with concepts that drift or change completely, is one of the core issues in IoT stream mining. This tutorial is a gentle introduction to mining IoT big data streams. The first part introduces data stream learners for classification, regression, clustering, and frequent pattern mining. The second part deals with scalability issues inherent in IoT applications, and discusses how to mine data streams on distributed engines such as Spark, Flink, Storm, and Samza.
Gianmarco De Francisci Morales is a scientist at QCRI. Previously he worked as a visiting scientist at Aalto University in Helsinki, as a research scientist at Yahoo Labs in Barcelona, and as a research associate at ISTI-CNR in Pisa. He received his PhD in computer science and Engineering from the IMT Institute for Advanced Studies of Lucca in 2012. His research focuses on scalable data mining, with an emphasis on Web mining and data-intensive scalable computing systems. He is an active member of the open source community of the Apache Software Foundation, working on the Hadoop ecosystem, and a committer for the Apache Pig project. He is one of the lead developers of Apache SAMOA, an open-source platform for mining big data streams. He commonly serves on the PC of several major conferences in the area of data mining, including WSDM, KDD, CIKM, and WWW. He coorganizes the workshop series on social news on the web (SNOW), colocated with the WWW conference.
Albert Bifet is an associate professor at Telecom ParisTech and honorary research associate at the WEKA Machine Learning Group at University of Waikato. Previously he worked at Huawei Noah's Ark Lab in Hong Kong, Yahoo Labs in Barcelona, University of Waikato and UPC BarcelonaTech. He is the author of Adaptive Stream Mining and Pattern Learning and Mining from Evolving Data Streams. He is one of the leaders of MOA and Apache SAMOA software environments for implementing algorithms and running experiments for online learning from evolving data streams. He is serving as cochair of the industrial track of IEEE MDM 2016, ECML PKDD 2015, and as cochair of BigMine (2015, 2014, 2013, 2012), and ACM SAC data streams track (2016, 2015, 2014, 2013, 2012).
Latifur Khan is a full professor (tenured) in the computer science department at the University of Texas at Dallas where he has been teaching and conducting research since September 2000. He received his PhD and MS degrees in computer science from the University of Southern California in August of 2000, and December of 1996 respectively. He has received prestigious awards including the IEEE Technical Achievement Award for intelligence and security informatics. Khan is an ACM distinguished scientist and a senior member of IEEE. He has chaired several conferences and serves (or has served) as associate editor on multiple editorial boards including IEEE Transactions on Knowledge and Data Engineering (TKDE) journal. He has conducted tutorial sessions in prominent conferences such as ACM WWW 2005, MIS2005, DASFAA 2007, and WI 2008 (Matching Words and Pictures - Problems, Applications, and Progress) and PAKDD 2011 (Data Stream Mining Challenges and Techniques).
Joao Gama received, in 2000, his PhD degree in computer science from the Faculty of Sciences of the University of Porto, Portugal. He joined the Faculty of Economy where he holds the position of associate professor. He is also a senior researcher and vice-director of LIAAD, a group belonging to INESC TEC. He has worked in several national and European projects on incremental and adaptive learning systems, ubiquitous knowledge discovery, learning from massive, and structured data, and others. He served as coprogram chair of ECML'2005, DS'2009, ADMA'2009, IDA' 2011, and ECM-PKDD'2015. He served as track chair on data streams with ACM SAC from 2007 – 2016. He organized a series of workshops on knowledge discovery from data streams with ECMLPKDD conferences and knowledge discovery from sensor data with ACM SIGKDD. He is author of several books in data mining (in Portuguese) and authored a monograph on knowledge discovery from data streams. He authored more than 250 peer-reviewed papers in areas related to machine learning, data mining, and data streams. He is a member of the editorial board of international journals ML, DMKD, TKDE, IDA, NGC, and KAIS.
Wei Fan is the head of Baidu Big Data Lab. He received his PhD in computer science from Columbia University in 2001. His main research interests and experiences are in various areas of data mining and database systems, such as, stream computing, high performance computing, extremely skewed distribution, cost-sensitive learning, risk analysis, ensemble methods, easy-to use nonparametric methods, graph mining, predictive feature discovery, feature selection, sample selection bias, transfer learning, time series analysis, bioinformatics, social network analysis, novel applications and commercial data mining systems. His coauthored paper received ICDM 2006 Best Application Paper Award, he led the team that used his random decision tree method to win 2008 ICDM Data Mining Cup Championship. He received 2010 IBM Outstanding Technical Achievement Award for his contribution to IBM Infosphere Streams. He is the associate editor of ACM Transaction on Knowledge Discovery and Data Mining (TKDD). At Huawei, he led his colleagues to develop Huawei StreamSMART, a streaming platform for online and real-time processing, query and mining of very fast streaming data. In addition, he also led his colleagues to develop a real-time processing and analysis platform of mobile broad band (MBB) data.
SA5: Computer Poker
Sam Ganzfried, Johannes Heinrich, and Kevin Waugh
Poker has been studied academically since the founding of game theory and in fact may have even been the inspiration for the field: the only motivating problem described in John Nash's PhD thesis, which defined and proved existence of the central solution concept, was actually a three-player poker game. Such interest has not waned over time. Last year when a computer program developed at Carnegie Mellon University competed against the strongest human two-player no-limit Texas hold 'em players in the inaugural Brains versus Artificial Intelligence Competition, thousands of poker fans followed with excitement. Earlier that year another monumental breakthrough was attained, as the two-player limit variation of Texas hold 'em (the smallest variant played competitively by humans) was essentially weakly solved (that is, an epsilon-Nash equilibrium was computed for such a small epsilon to be statistically indistinguishable from zero in a human lifetime) by researchers at the University of Alberta. This result was published in Science. Poker, and particularly Texas hold 'em, is tremendously popular for humans, and online poker is a multibillion dollar industry. Computer poker has proved to be one of the most visible applications of research in computational game theory.
Prerequisite knowledge: No prerequisite knowledge is needed.
Sam Ganzfried is an assistant professor in computer science at Florida International University. He received a PhD in computer science from Carnegie Mellon in 2015. He created two-player no-limit Texas hold 'em agent Claudico that competed in the Brains versus Artificial Intelligence Competition against the strongest humans in the world.
Johannes Heinrich is completing his PhD in computer science at University College London, researching reinforcement learning from self-play in imperfect-information games His PhD was supported by DeepMind, where he worked on game-theoretic approaches to reinforcement learning. He developed SmooCT, an MCTS-based agent that won 3 silver medals at the ACPC.
Kevin Waugh is a research scientist at Facebook. He has been involved with the computer poker competition since 2008, competing while obtaining his MSc at University of Alberta under Michael Bowling. Subsequently, he studied under Drew Bagnell at Carnegie Mellon University. He has chaired the competition since 2014.
SP1: Recent Advances in Distributed Machine Learning
Tie-Yan Liu, Wei Chen, and Taifeng Wang
In recent years, artificial intelligence has demonstrated its power in many important applications. Besides the novel machine learning algorithms (for example, deep neural networks), their distributed implementations play a very critical role in these successes. In this tutorial, we will first review popular machine learning models and their corresponding optimization techniques. Second, we will introduce different ways of parallelizing machine learning algorithms, that is, data parallelism, model parallelism, synchronous parallelism, asynchronous parallelism, and so on, and discuss their theoretical properties, advantages, and limitations. Third, we will discuss some recent research works that try to overcome the limitations of standard parallelization mechanisms, including advanced asynchronous parallelism and new communication and aggregation methods. Finally, we will introduce how to leverage popular distributed machine learning platforms, such as Spark MlLib, DMTK, Tensorflow, to parallelize a given machine learning algorithm, in order to give the audience some practical guidelines on this topic.
Prerequisite knowledge: This tutorial is accessible to both academic researchers and industrial practitioners, especially for those who are working on/around deep learning and/or large-scale machine learning. Only very preliminary knowledge to machine learning, deep learning, and optimization are required.
Tie-Yan Liu is a principal researcher of Microsoft Research Asia, an adjunct professor of Carnegie Mellon University, and an honorary professor of Nottingham University. His research interests include machine learning, information retrieval, algorithmic game theory, and others. He is a senior member of the IEEE and the ACM.
Wei Chen is a researcher in Machine Learning Group, Microsoft Research Asia. She is interested in distributed machine learning, deep learning theory, etc. She has published tens of papers at AAAI, NIPS, IJCAI, and other conferences. She obtained her PhD in mathematics from the Chinese Academy of Sciences in 2011.
Taifeng Wang is a lead researcher in the machine learning group, Microsoft Research Asia. His research interests include machine learning, distributed system, search ads click prediction, graph mining, and others. Currently, he is leading Microsoft's open source project DMTK (distributed machine learning toolkit), which already acquired more than 1,600 stars on Github.
SP2: Statistical Relational Artificial Intelligence: Logic, Probability and Computation
Luc De Raedt, David Poole, Kristian Kersting, and Sriraam Natarajan
Both predicate logic and probability theory extend propositional logic, one by adding relations, individuals and quantified variables, the other by allowing for measures over possible worlds and conditional queries. While logical and probabilistic approaches have often been studied and used independently within artificial intelligence, they are not in conflict with each other but they are synergistic. Indeed, there has been a considerable body of research in combining first-order logic and probability in the area known as statistical relational artificial intelligence (StarAI).
Relational probabilistic models — we use this term in the broad sense, meaning any models that combine relations and probabilities — form the basis of StarAI, and can be seen as combinations of probability and predicate calculus that allow for individuals and relations as well as probabilities. In building on top of relational models, StarAI goes far beyond reasoning, optimization, learning and acting optimally in terms of a fixed number of features or variables, as it is typically studied in machine learning, constraint satisfaction, probabilistic reasoning, and other areas of AI. This tutorial provides a gentle introduction to the state of the art and the current challenges in StarAI.
Luc De Raedt is a full professor at KU Leuven (Belgium) and holder of an ERC Advanced Grant. His research interests are in AI, machine learning, data mining and uncertainty reasoning. He is working on probabilistic programming, the integration of constraint programming with data mining, and the automation of data science.
David Poole is a professor of computer science at the University of British Columbia. He is a coauthor of two AI textbooks. He is a Fellow of the Association for the Advancement Artificial Intelligence (AAAI) and the winner of the Canadian AI Association (CAIAC) 2013 Lifetime Achievement Award.
Kristian Kersting is an associate professor of computer science at the TU Dortmund University. He is a coauthor of a textbook on statistical relational AI, received the ECCAI Dissertation Award in 2006, cochaired ECML PKDD 2013 and will cochair UAI 2017.
Sriraam Natarajan is an associate professor of computer science and informatics at Indiana Univeristy. He is a coauthor of a textbook on statistical relational AI, received the ARO young investigator award and has cochaired the first five workshops on statistical relational AI.
SP3: AI Planning for Robotics
Daniele Magazzeni and Michael Cashmore
This tutorial provides an overview of aAI planning and scheduling (P&S) for robotics domains. We will discuss how P&S formalisms and tools are used in robotics domains; examine several case studies; and provide an overview of ROSPlan, a framework for integrating PDDL P&S with ROS. The tutorial aims to discuss the main challenges related to planning for autonomous robots within the robotics community (state estimation, deliberative, reactive, continuous planning and execution, and others.) By so doing, we hope to encourage the rapid development of P&S techniques for robotics.
The main goal of this tutorial is to show how state-of-the-art formalisms in planning and scheduling (P&S) are used to model robotics domains; some of the challenges and solutions involved in dispatching plans on-board robotic platforms; and to provide an overview of the ROSPlan framework.
Prerequisite knowledge: The target audience is the broad AI community (including researchers in robotics). General knowledge of artificial intelligence concepts such as search and optimization is presumed.
Daniele Magazzeni leads the AI Planning group at King's College London. His research explores the links between AI planning and model-checking, with a focus on hybrid systems and robotics. He is the organizer of the Dagstuhl Workshop on AI Planning and Robotics to be held in 2017.
Michael Cashmore is a research associate at King's College London. His research explores integration between planning and autonomous robots, and planning in mixed discrete-continuous domains. He is the developer of ROSPlan; and SMTPlan, a PDDL+ planner for hybrid systems.
SP4: Modeling and Solving AI Problems in Picat
Roman Barták and Neng-Fa Zhou
Picat is a logic-based multiparadigm language that integrates logic programming, functional programming, constraint programming, and scripting. Picat takes many features from other languages, including logic variables, unification, backtracking, pattern-matching rules, functions, list/array comprehensions, loops, assignments, tabling for dynamic programming and planning, and constraint solving with CP (constraint programming), SAT (satisfiability), and MIP (mixed integer programming).
The goal of this tutorial is to show how to declaratively model some classical AI problems, including constraint satisfaction and planning problems, and solve them efficiently using Picat. The tutorial is targeted toward researchers who are interested in AI algorithms and need a high-level modeling tool for rapid prototyping and efficient solving of AI problems. The audience will learn the basics of the Picat language and its programming, and some general techniques and tricks for modeling constraint satisfaction and planning problems. Modeling has been an important topic in constraint programming and operations research. However, modeling in planning has received little attention. The major novelty of the tutorial is to show how declarative and concise models in Picat can lead to efficient solutions. There will also be sessions explaining how high-level models are compiled to encodings for fast solving by existing solvers.
Roman Barták works as a full professor and a researcher at Charles University in Prague (Czech Republic), where he leads Constraint Satisfaction and Optimization Research Group. His research work focuses on techniques of constraint satisfaction and modeling and their application to planning, scheduling, and other areas.
Neng-Fa Zhouis a professor of computer and information science at CUNY Brooklyn College and Graduate Center. He is the principal creator of the Picat system. He is an enthusiast of problem solving methods and tools. He won prizes in several competitions, including CSP, ASP, and MiniZinc Challenge.
SP5: AI for Data-Driven Decisions in Water Management
Biplav Srivastava and Sandeep S. Sandha
Water is unique in its role as a life preserver. It is important to all members of a society, who can benefit with decision support aids (that is, AI systems) which can help them understand water pollution data and alternative decision choices that they may have.
The aim of the tutorial is to make both early and experienced researchers aware of the water management area, its issues and opportunities for using AI techniques to make real world impact that matters. In particular, we discuss available open water pollution data, both quantitative and qualitative in nature from a modality of sensors, that researchers can use to help a wide spectrum of decision makers, for example, farmers, tourists, environmentalists, health officials, policy makers and businesses. We ground the presentation based on our work in (1) using open data released by government agencies, (2) developing data platform and mobile apps, (3) developing use-cases benefiting from pollution data like tourism and industry inspection, and (4) field experiments at polluted rivers in India like Yamuna at Delhi, Hindon at Meerut and Ganga at Haridwar with multiple sensors for these use-cases. Among systems referenced, BlueWater is a data platform to collect, reconcile and share quantitative pollution data from real-time sensors and historical data, and powers the released GangaWatch mobile app to make sense from all. Since mobile phones are omnipresent, mobile apps like CreekWatch on iOS and Neer Bandhu (Water Friend) on Android enable people to share qualitative assessment of water quality and validate physical sensing. The collected data is released to community via APIs.
Prerequisite knowledge: The tutorial will expect the audience to have a basic computer science background consisting of data structures, algorithms, databases and search.
Biplav Srivastava is a research staff member and master inventor at IBM Research, Yorktown Heights and an ACM distinguished scientist and distinguished speaker. His research deals with enabling people to make rational decisions despite real world complexities of poor data, changing goals and limited resources. With expertise in artificial intelligence, services and sustainability over the last two decades, he has been working on AI-driven innovations for the last 6 years in smart city areas of water and traffic with global relevance.
Sandeep Sandha was a research software engineer at IBM Research, India from 2014–2016 where he spearheaded software development for BlueWater and the GangaWatch apps. He has worked extensively on IoT and data management for water, natural sciences and health. A BTech from IIT Rookee, India, he is in the doctoral program at the University of California, Los Angeles.
SP6: Social Data Bias in Machine Learning: Impact, Evaluation, and Correction
Fred Morstatter and Huan Liu
The goal of this tutorial is to introduce how data issues can lead to inaccurate models and research conclusions. These biases arise across a wide swath of artificial intelligence tasks, from biases in computer vision to social media data. These biases lead to machine learning models that make spurious predictions. Using social media as an example, these biases can stem from the make up of user base (demographic bias), as well as from malicious users who try to manipulate the discussion on the site (malicious user bias). Sampling strategies from the sites themselves lead to data that is not based on human activity (sample bias). Furthermore, these sampling strategies can be manipulated to tamper with research performed from these data outlets (malicious sample bias). In this tutorial we will discuss how these representativity issues can lead to misinformed conclusions based upon data from social media, as well as incorrect machine learning models. We will demonstrate case studies of how the results of these biased data sources cause fundamental errors in data analysis. Finally, we will present approaches for correcting for this bias, enabling increased credibility and generalizability of research.
Fred Morstatter is a PhD student at Arizona State University. Among his publications are several papers which investigate bias and representativity in streaming social media data, as well as book on mining Twitter: Twitter Data Analytics.
Huan Liu is a professor of computer science and engineering at Arizona State University. His research interests are in social computing, data mining, machine learning, and artificial intelligence. He is an IEEE Fellow.
SUA1: Deep Learning Implementations and Frameworks
Seiya Tokui, Kenta Oono, and Atsunori Kanemura
This tutorial is for AI researchers and practitioners who want to utilize deep learning to develop AI systems or systems making use of AI technologies. It is tailored to help them to choose software frameworks suitable for their applications from various candidates including Caffe and TensorFlow. Choosing an appropriate framework can improve productivity and increase the utility and popularity of work. In this tutorial, the audience will learn guidelines for selecting an appropriate one with the understandings of its features.
This tutorial uniquely discusses design principles that drive the development of deep learning frameworks and shows the pros and cons of adopting each principle or not. Deciding which principle to adopt poses a tradeoff and makes differences for resulting deep learning systems in for example, processing speed, debugging easiness, and the efficiency in dynamic model changes. We examine coding examples for TensorFlow, Keras, and Chainer to give deep understanding of their internal mechanisms, as well as their usage.
Prerequisite knowledge: This tutorial provides background knowledge on neural networks to be self-contained. However, we assume that the audience know the basic concepts of machine learning, especially supervised learning. The experience of using neural networks and the Python programming language is recommended.
Seiya Tokui is a researcher at Preferred Networks, Inc. and a Ph.D student at the University of Tokyo, from which he received a master's degree in mathematical informatics in 2012. He is the lead developer of Chainer, a deep learning framework. His research interests include deep learning and generative models.
Kenta Oono is an engineer at Preferred Networks, Inc., Japan. He received his master's degree in mathematics at the University of Tokyo in 2011. He is a core developer of a deep learning framework, Chainer. His research interest is on deep learning, bioinformatics, and theoretical analysis of machine learning.
Atsunori Kanemura is a research scientist at Mathematical Neuroinformatics Group and Machine Learning Team, National Institute of Advanced Industrial Science and Technology (AIST), Japan. He obtained the PhD degree in informatics from Kyoto University in 2009. His research interests include machine learning, statistical signal processing, and analysis of human data.
SUA2: Learning Bayesian Networks for Complex Relational Data
Oliver Schulte and Ted Kirkpatrick
Many organizations maintain critical data in a relational database. The tutorial describes the statistical and algorithmic challenges of constructing models from such data, compared to the single-table data that is the traditional emphasis of machine learning. We extend the popular model of Bayesian networks to relational data, integrating probabilistic associations across all tables in the database. Extensive research has shown that such models, accounting for relationships between entities (such as actors featured in a movie, or ratings of movies by viewers), have greater predictive accuracy than models learned without relational information. We illustrate the challenges in learning such models, and their solutions, in a real-life running example, the Internet Movie Database. The tutorial shows how Bayesian networks support several relational learning tasks, such as querying multirelational frequencies, link-based classification, and relational anomaly detection. We describe how standard machine learning concepts can be extended and adapted for relational data, such as sufficient statistics, model likelihood, model selection, and statistical consistency.
Prerequisite knowledge: The tutorial is self-contained as it addresses both researchers with a background in machine learning, but possibly none in logic, and researchers with a background in logic, but possibly none in machine learning. The current version of slides is available on the supplemental tutorial website.
Oliver Schulte is a professor in computing science at Simon Fraser University, Vancouver, Canada. (PhD Carnegie Mellon 1997). He has published papers in leading AI and machine learning venues on a variety of topics, including relational learning, learning Bayesian networks, game theory, and scientific discovery.
Ted Kirkpatrick is an associate professor of computing science at Simon Fraser University. With Oliver Schulte, he has coauthored articles on modeling relational data via Bayesian networks. Previously, he studied statistical design space exploration for human computer interaction.
SUA3: Causal Inference and the Data-Fusion Problem
Machine Learning is usually dichotomized into two categories, passive (such as supervised learning) and active (such as reinforcement learning) which, by and large, are studied separately. Reality is more demanding. Passive and active modes of operation are but two extremes of a rich spectrum of data-collection modes (also called research designs) that generate the bulk of the data available in practical, large scale situations. In typical medical explorations, for example, data from multiple observations and experiments are collected, coming from distinct experimental setups, different sampling conditions, and heterogeneous populations. Similarly, in a more basic setting, a baby learns from its environment by both passively observing others and interacting with its environment by actively performing interventions.
In this tutorial, we will review concepts, principles, and mathematical tools that were found useful in reasoning about passive and active modes of interaction, how these modes relate to causal and counterfactual reasoning, and how these results have been used in data-intensive sciences. In particular, we will introduce the data-fusion problem, which is concerned with piecing together multiple datasets collected under heterogeneous conditions (to be defined) so as to obtain valid answers to queries of interest. The tutorial will include discussions on issues of confounding control, policy analysis, misspecification tests, heterogeneity, selection bias, and reinforcement learning.
The following topics will be emphasized:
- The 3-layer causal hierarchy: association, intervention, and counterfactuals. [Reference]
- What mathematics can tell us about "transfer learning" or "generalizing across domains" [Reference] [Reference]
- What causal analysis tells us about recovery from selection bias. [Reference] [Reference]
- The relationship between counterfactual inference and reinforcement learning. [Reference] [Reference]
For additional material please see the tutorial supplementary website.
Elias Bareinboim is an assistant professor in the Department of Computer Science at Purdue University. His research focuses on causal and counterfactual inference and their applications to data-driven fields. Bareinboim received a Ph.D. in Computer Science from UCLA advised by Judea Pearl. His doctoral thesis was the first to propose a general solution to the problem of "data-fusion" and to provide practical methods for combining datasets generated under different experimental conditions. Bareinboim's recognitions include IEEE AI's 10 to Watch, the Dan David Prize Scholarship, the Yahoo! Key Scientific Challenges Award, and the 2014 AAAI Outstanding Paper Award.
SUA4: Eliciting High-Quality Information
Boi Faltings and Goran Radanovic
AI systems often depend on information provided by other agents, for example sensor data or crowdsourced human computation. Usually, providing accurate and relevant data requires costly effort that agents may not always be willing to provide. Thus, it becomes important both to verify the correctness of data, but also to provide incentives so that agents that provide high-quality data are rewarded while those that do not are discouraged by low rewards.
We will cover different settings and the assumptions they admit, including sensing, human computation, peer grading, reviews and predictions. We will survey different incentive mechanisms, including proper scoring rules, prediction markets and peer prediction, Bayesian truth serum, peer truth serum, and the settings where each of them would be suitable. As an alternative, we also consider reputation mechanisms. We complement the game-theoretic analysis with practical examples of applications in prediction platforms, community sensing and peer grading.
This tutorial is intended for AI researchers and practitioners working with contributed or crowdsourced data, be it in machine learning, AI on the web, sensing and computational sustainability, or optimization.
Boi Faltings is a full professor at the Swiss Federal Institute of Technology in Lausanne (EPFL) and has been one of the pioneers on the topic of mechanisms for truthful information elicitation, with the first work dating back to 2003. He is a fellow of AAAI and EurAI.
Goran Radanovic is a Post-doctoral fellow at Harvard University. He received his PhD from the Swiss Federal Institute of Technology on the topic of mechanisms for information elicitation. His work has been widely published at AAAI, AAMAS and journals.
SUA5: Discrete Sampling and Integration for the AI Practitioner
Supratik Chakraborty, Kuldeep S. Meel, and Moshe Y. Vardi
Discrete sampling and integration are fundamental problems in Artificial Intelligence, with a wide variety of applications spanning probabilistic inference, statistical machine learning, network reliability analysis and the like. This tutorial will introduce the audience to the extensive theoretical and practical research work done in this area over the past few decades. We will discuss in detail the promising recent approach of using universal hashing to integrate and sample with strong formal guarantees, while also scaling to large and realistic problem sizes.
Supratik Chakraborty is the Bajaj chair professor in the Department of computer science and Engineering at Indian Institute of Technology Bombay, India. Supratik's current research interests include constrained sampling and counting techniques and their applications, and also formal methods for analyzing hardware, software and biological systems.
Kuldeep Meel is a final year PhD student at Rice University. He is the recipient of 2016-17 IBM PhD Fellowship and Lodieska Stockbridge Vaughn Fellowship. His research won best student paper award at International Conference on Constraint Programming 2015 and 2014 Vienna Center of Logic and Algorithms Outstanding Masters thesis award.
Moshe Y. Vardi is the George Distinguished Service Professor in Computational Engineering and director of the Ken Kennedy Institute for Information Technology at Rice University. He is the author and coauthor of over 500 papers, as well as two books. He is a member of several honorary academies.
SUP1: Interactive Machine Learning: From Classifiers to Robotics
Brad Hayes, Ece Kamar, and Matthew E. Taylor
To make virtual agents and physical robots solve real-world tasks, it often becomes necessary to learn not only from static datasets or simulated oracles, but directly from humans. Unfortunately, some of the assumptions underlying traditional statistical machine learning approaches become invalid when learning from data provided by slow, inaccurate, or inconsistent trainers. Furthermore, many additional considerations that are typically outside the purview of machine learning experts, such as user interface, become critical. This tutorial will (1) survey selected existing work in this exciting and growing field; (2) propose a framework to classify and understand different types of work in this area, as well as highlight important opportunities for additional work; and (3) cover a selection of practical considerations, such as participant recruitment and compensation, and useful toolkits or testbeds.
Prerequisite knowledge: The majority of the material in this tutorial will be understandable without any domain expertise. There will sections in which having a background in machine learning at the level of a one-semester graduate class will be useful for understanding algorithmic details.
Brad Hayes is a postdoctoral associate in the Interactive Robotics Group at MIT. Focusing on enabling fluent human-robot collaboration and interpretable machine learning, his work develops the algorithms necessary to build capable, supportive, and interactive autonomous robotic systems that operate safely and legibly around humans.
Ece Kamar is a researcher at the Adaptive Systems and Interaction group at Microsoft Research Redmond. Ece earned her PhD in computer science from Harvard University. She works on a number of subfields of AI; including planning, machine learning, multiagent systems and human-computer teamwork.
Matthew E. Taylor is an assistant professor at Washington State University and holds the Allred Distinguished Professorship in Artificial Intelligence. His group, the Intelligent Robot Learning Lab, researches topics including intelligent agents, multiagent systems, reinforcement learning, transfer learning, and robotics.
SUP2: Knowledge Graph Construction from Text
Jay Pujara, Sameer Singh, and Bhavana Dalvi
With the proliferation of large collections of unstructured text, the problem of extracting structured knowledge and integrating it into a coherent knowledge graph has become increasingly important. From Siri to summarization, knowledge graphs have become an integral component of many applications. Due to its importance, this area has been an active area of research spanning areas of natural language processing, information extraction, information integration, databases, search, and machine learning.
In this tutorial, we will present an accessible and structured overview of the existing approaches for extracting candidate facts from text and incorporating these into a well-formed knowledge graph. The first half of this tutorial will cover knowledge extraction, with a focus on the underlying NLP tasks and successful approaches for converting text into candidate facts. The second half of the tutorial will focus on machine learning approaches such as tensor factorization, deep learning, probabilistic graphical models, and random walk strategies for integrating candidate facts into a complete and coherent knowledge graph.
Our approach includes identifying the common themes and challenges in the area, and comparing and contrasting the existing approaches on the basis of these aspects. We will highlight the key applications, useful datasets, and prominent tools for knowledge graph construction, and offer practical advice. We believe our unifying framework will provide the necessary tools and perspectives to enable the newcomers to the field to explore, evaluate, and develop novel techniques for automated knowledge graph construction.
Prerequisite knowledge: The tutorial is designed to be self-contained, but familiarity with the basic concepts from artificial intelligence and related fields will be helpful.
Jay Pujara is a postdoctoral researcher at the University of California, Santa Cruz. He has won best paper awards for his work on knowledge graph identification and entity resolution for knowledge graphs, and recently coorganized the Automated Knowledge Base Construction (AKBC) workshop. He received his PhD from the University of Maryland College Park.
Sameer Singh is an assistant professor at the University of California, Irvine. His research interests are in interactive and large-scale machine learning applied to information extraction and natural language processing. He was a postdoc at University of Washington, and got his PhD from the University of Massachusetts Amherst.
havana Dalvi is a research scientist at the Allen Institute for Artificial Intelligence. She has worked in the area of information extraction and machine learning for the past seven years. She earned her PhD from Carnegie Mellon University in 2015 and received a Google PhD fellowship in information extraction from 2013–2015.
SUP3: Introduction to multiAgent Path Finding
Ariel Felner, Sven Koenig, and Glenn Wagner
Teams of agents often have to assign target locations among themselves and then plan collision-free paths to their target locations. Examples include automated warehouse systems, office robots and game characters in video games. The resulting multiAgent Path Finding (MAPF) problem is NP-hard to solve optimally and even NP-hard to approximate within any constant factor less than 4/3. It has been studied in artificial intelligence, robotics and theoretical computer science.
The tutorial will describe different variants of the MAPF problem (including their complexity), different approaches to solve these variants and their applications, including considerations for using them on real robots. The tutorial will cover a variety of optimal approaches (including dedicated algorithms and reductions to other well-studied problems, such as satisfiability), suboptimal approaches and fast but incomplete approaches.
Prerequisite knowledge: Participants are only expected to be familiar with standard single-agent path-finding techniques (such as A* and RRTs).
Ariel Felner is an associate professor at Ben-Gurion University. He is the chair of the Israeli Association for Artificial intelligence (IAAI) and a council member SoCS. He is interested in all aspects of heuristic search with specific care in pedagogical and historical aspects of teaching concepts in this field.
Sven Koenig is a professor in computer science at the University of Southern California. Most of his research centers around techniques for decision making that enable single agents (such as robots) and teams of agents to act intelligently in their environments. See idm-lab.org for more information.
Glenn Wagner is a post-doctoral fellow at Carnegie Mellon University. He received a BS in mechanical engineering from the California Institute of Technology in 2009 and a PhD in robotics from Carnegie Mellon university in 2015. His research explores multiagent path finding and analysis of high-dimensional systems.
SUP4: Predicting Human Decision-Making: Tools of the Trade
Ariel Rosenfeld and Sarit Kraus
Human decision-making often transcends our formal models of rationality. Designing intelligent agents that interact proficiently with people necessitates the modeling of human behavior and the prediction of their decisions.
In this three and a half hour tutorial, we will focus on the prediction of human decision-making and its use in designing intelligent human-aware automated agents of varying natures; from purely conflicting interaction settings (for example, security and games) to fully cooperative interaction settings (for example, advise provision, human rehabilitation). We will present computational representations, algorithms and empirical methodologies for meeting the challenges that arise from the above tasks in both a single interaction (one-shot) and repeated interaction settings. The tutorial will also review recent advances, current challenges and future directions for the field
Prerequisite knowledge: In the course of the tutorial we will present techniques and ideas using machine learning, game-theoretical and general AI concepts. The basis for these concepts will be covered as part of the tutorial; however, a basic familiarity with the above concepts is encouraged.
Ariel Rosenfeld is currently completing his PhD in computer science at Bar-Ilan University, Israel. He obtained a BSc in computer science and economics, magna cum laude, from Tel-Aviv University, Israel. Rosenfeld's research focus is human-agent interaction and he has published at top venues such as AAAI, IJCAI, AAMAS, and ECAI.
Sarit Kraus is a professor of computer science at Bar-Ilan University, Israel and an adjunct professor at the University of Maryland. She has focused her research on intelligent agents and multiagent systems. Kraus has published over 300 papers in leading journals and major conferences and has presented many invited talks and tutorials.
SUP5: Neuroevolution Reinforcement Learning
Neuroevolution RL will first be motivated and compared with value-function RL, and shown to be strong in domains with large state and action spaces and where the state of the world is not fully known. Basic neuroevolution techniques will be reviewed and illustrated with state of the art applications in robotics, game playing, intelligent agents, and multiagent systems. Future opportunities will then be reviewed, focusing in particular on generative and developmental approaches, and the construction of deep learning architectures, where the structure and hyperparameters of the network are evolved and the weights are trained.
Risto Miikkulainen is a professor of computer science at the University of Texas at Austin and a Fellow at Sentient Technologies, Inc. He is the author of more than 370 articles on neuroevolution, connectionist natural language processing, and the computational neuroscience of the visual cortex. He is an associate editor of Cognitive Systems Research and IEEE Transactions on Computational Intelligence and AI in Games, a member of the Board of Governors of the International Neural Networks Society, and a Fellow of the IEEE.
SUP6: Artificial Intelligence and Video Games
AAAI-17 Call for Papers
Special Track on Cognitive Systems
Special Track on Computational Sustainability
Special Track on Integrated Systems
EAAI Symposium Call
IAAI-17 Call for Papers
Tutorial Forum Call
Student Abstract Call
DC Call for Applications
Senior Member Track Call