Dr. Sven Koenig

Title: Planning for Large-Scale Multi-Robot Systems

Bio: Sven Koenig is a professor in computer science at the University of Southern California (USC). He received his Ph.D. degree in computer science from Carnegie Mellon University and holds M.S. degrees from the University of California at Berkeley and Carnegie Mellon University. Most of Sven's research centers around techniques for decision making (planning and learning) that enable single situated agents (such as robots or decision-support systems) and teams of agents to act intelligently in their environments and exhibit goal-directed behavior in real-time, even if they have only incomplete knowledge of their environment, imperfect abilities to manipulate it, limited or noisy perception or insufficient reasoning speed. He believes that finding good solutions to these problems requires approaches that cut across many different fields. Sven is a fellow of the Association for the Advancement of Artificial Intelligence (AAAI) and an ACM Senior Member. He is the recipient of research awards (including an NSF CAREER Award, an IBM Faculty Partnership Award, a Charles Lee Powell Foundation Award and a Raytheon Faculty Fellowship Award) and educational awards (including a Mellon Mentoring Award and an SAIC Student Advisement Award). He was also a program director at the National Science Foundation (NSF) from 2010 to 2012, where he received the Director's Award for Collaborative Integration in both 2011 and 2012. Several of his students won awards as well (including best paper, best dissertation, best research assistant and best teaching assistant awards) and took their first jobs in academia. Sven has edited several conference proceedings and published more than 150 papers in various areas of artificial intelligence and robotics.

Abstract: Multi-robot systems are now being used in industry. For example, hundreds of robots navigate autonomously in Amazon fulfillment centers to move inventory pods all the way from their storage locations to the inventory stations that need the products they store (and vice versa). Autonomous aircraft towing vehicles will soon tow aircraft all the way from the runways to their gates (and vice versa), thereby reducing pollution, energy consumption, congestion, and human workload. Path planning for these robots is difficult, yet one must find high-quality collision-free paths for them in real-time. Shorter paths result in higher throughput or lower operating costs (since fewer robots are required). I will discuss different versions of such multi-robot path-planning problems, algorithms for solving them, and their applications. I will also discuss a planning architecture that combines ideas from artificial intelligence and robotics. It makes use of a simple temporal network to post-process the output of a multi-robot path-planning algorithm in polynomial time to create a plan-execution schedule for robots that provides a guaranteed safety distance between them and exploits slack to absorb imperfect plan executions and avoid time-intensive re-planning in many cases. This talk is suitable for audiences with some computer science background. A background in artificial intelligence or robotics is not necessary.

Dr. Carlos Lopes-Molina

Title: Zadeh and Marr: A tale of two pioneers

Abstract: This talks draws parallelisms between two pioneers in their respective fields: Lotfi Zadeh and David Marr. Zadeh, in his quest of producing intelligent machines, developed a theory of thinking based on human like representation. Marr, on his account, had the opposive quest of reverse-engineering the most relevant thinking engine ever analyzed: the human brain. In this talk we review the convergences and divergences in the ideas of both pioneers, as well as those in the fields created from their work. This will unravel in a common study of how certain topics, as information representation and tolerance to ambiguity, became central to both theories. As a conclusion, the talk will question the actual overlap of the fields of Fuzzy Set Theory and Neurosciences, and the possibilities to exchange developments in both directions.

Dr. Felipe Meneguzzi

Title: Goal Recognition with Real World Data

Bio: Felipe Meneguzzi is a professor in computer science at the Pontifical Catholic University of Rio Grande do Sul (PUCRS). He received his PhD degree in Computer Science from King's College London (KCL) and holds an MSc degree from PUCRS. Prior to hs current appointment Felipe was a Research Fellow at Carnegie Mellon University (CMU) and prior to his academic career he was a Technical Leader at Hewlett Packard Brazil. Felipe's research focuses on reasoning mechanisms for autonomous agents, with a focus on integrating various approaches to achieve not only efficient but also socially good behaviour by following societal norms. To this end, Felipe has contributed with approaches involving planning, machine learning and logic spanning the width of modern AI research. Felipe is currently classified as a highly productive research in Brazil (PQ2) and received numerous awards for his work (including the 2016 Google Research Award for Latin America), service (including the best PC member for PRIMA 2013), and for his students. He currently leads the Research Group on Autonomous Systems at PUCRS where he carries out research on Goal Recognition, Automated Planning and Deep Learning.

Abstract: Plan and goal recognition is the task of inferring the plan and goal of an agent through the observation of its actions and its environment and has a number of applications on computer-human interaction, assistive technologies and surveillance. Recent work on such techniques using planning domain theories require almost perfect engineering of the domain theory and usually rely on assumptions of observability being either full, noise-free, or both. These assumptions are too strong in the real world, regardless of the technique used to translate sensor data into symbolic logic-based observations. In this talk, we discuss plan recognition techniques based on classical planning domain theories gradually relax all of these assumptions by leveraging advances on both symbolic planning and machine learning.

Dr. Hendrik Blockeel

Title: Multi-directional ensembles of decision trees

Bio: Hendrik Blockeel is a full professor and vice-chair at the Department of Computer Science of KU Leuven, and a member of the DTAI research lab (Declarative Languages and Artificial Intelligence). His research interests lie mostly within machine learning, data mining, and AI-based modeling, with particular interests in decision tree based approaches and probabilistic-relational learning. He has published over 150 peer-reviewed articles on these topics, has given keynote talks at multiple international conferences, and has chaired or organized several international conferences and schools. He has taught courses on computer science, machine learning, and statistics at the Universities of Leuven and Leiden, as well as at several international seasonal schools. He is a member of the Informatics board of the Lorentz Center in Leiden; action editor for the journals Machine Learning and Data Mining and Knowledge Discovery; and publications chair for ECML/PKDD, Europe’s premier conference series on machine learning and knowledge discovery. He was awarded Fellowship by EurAI, the European Association for Artificial Intelligence, in 2015.

Abstract: Ensembles of decision trees are among the most frequently used predictive models, with Random Forests and Gradient Boosted Trees as their most successful representatives. In this talk, I will discuss another instantiation: multi-directional ensembles of decision trees, as implemented in the MERCS approach. In a multi-directional ensemble, different trees may have different targets, so that the ensemble as a whole can be used to predict any variable from any other variable. This makes MERCS models very flexible in terms of the tasks they can perform. At the same time, they inherit the efficiency, accuracy, versatility and ease-of-use of decision tree ensembles. I will discuss the advantages and challenges of learning MERCS models and illustrate how they can be used for a variety of tasks.