Gautam Biswas is a Professor of Computer Science, Computer Engineering, and Engineering Management in the EECS Department and a Senior Research Scientist at the Institute for Software Integrated Systems (ISIS) at Vanderbilt University. He has an undergraduate degree in Electrical Engineering from the Indian Institute of Technology (IIT) in Mumbai, India, and M.S. and Ph.D. degrees in Computer Science from Michigan State University in E. Lansing, MI.

Prof. Biswas conducts research in Intelligent Systems with primary interests in hybrid modeling, simulation, and analysis of complex embedded systems, and their applications to diagnosis, prognosis, and fault-adaptive control. As part of this work, he has worked on fault-adaptive control of fuel transfer systems for aircraft, and Advanced Life Support systems for NASA. He has also initiated new projects in health management of complex systems, which includes online algorithms for distributed monitoring, diagnosis, and prognosis. More recently, he is working on data mining for diagnosis, and developing methods that combine model-based and data-driven approaches for diagnostic and prognostic reasoning. This work, in conjunction with Honeywell Technical Center and NASA Ames, includes developing sophisticated data mining algorithms for extracting causal relations amongst variables and parameters in a system. In other research projects, he is involved in developing simulation-based environments for learning and instruction. The most notable project in this area is the Teachable Agents project, where students learn science by building causal models of natural processes. He has also developed innovative educational data mining techniques for studying students’ learning behaviors and linking them to metacognitive strategies. His research has been supported by funding from NASA, NSF, DARPA, and the US Department of Education. His industrial collaborators include Airbus, Honeywell Technical Center, and Boeing Research and Development. He has published extensively, and has over 300 refereed publications.

Dr. Biswas is an associate editor of the IEEE Transactions on Systems, Man, and Cybernetics, Prognostics and Health Management, and Educational Technology and Society journal. He has served on the Program Committee of a number of conferences, and most recently was Program co-chair for the 18th International Workshop on Principles of Diagnosis and Program co-chair for the 15th International Conference on Artificial Intelligence in Education. He is currently serving on the Executive committee of the Asia Pacific Society for Computers in Education and is the IEEE Computer Society representative to the Transactions on Learning Technologies steering committee. He is also serving as the Secretary/Treasurer for ACM Sigart. He is a senior member of the IEEE Computer Society, ACM, AAAI, and the Sigma Xi Research Society.

Self-Improving systems typically use machine learning techniques to learn new knowledge and procedures in ways that make them better in performing the tasks they are designed for. Self-Improving methods may be applied to intelligent tutors along a number of different dimensions. Intelligent tutors contain three primary modules: (1) Domain module, which typically contains the knowledge of the domain and tasks the users are supposed to learn; (2) Learner module that keeps track of what users have learned, and where they have made errors; and (3) Pedagogical module that contains information about the approach the system takes to help the user learn the required content. This may include tutorial suggestions, informing the users about their progress, and determining the activities users should perform as part of their learning tasks. A self-improving tutor may apply machine learning techniques to data and information collected from users, instructors, and domain experts to improve upon one or more of its three primary modules. In my presentation, I will discuss machine learning methods that we have applied to improve the learner modeling module of tutoring systems. Our tutors are specifically designed to be open-ended learning environments (OELEs), where the user is presented with problems in a specific domain, and a set of tools and resources that facilitate learning and problem solving in that domain (Biswas, et al., 2016; Segedy, et al., 2015). Users have choice in their approach to problem solving, which in addition to learning domain content also facilitates learning of problem solving strategies and metacognitive processes. Therefore, our learner models not only track users domain knowledge, but also their learning and problem solving processes as they go about their assigned tasks on the system. In our work, we have used a combination of coherence measures (Segedy, et al., 2015) and exploratory sequence mining methods (Kinnebrew, et al., 2013; 2017) to understand users learning and problem solving behaviors. We have developed offline unsupervised learning methods to characterize users into categories of learners (Segedy, et al., 2015; Zhang & Biswas, 2016), and developed methods to combine coherence measures and derived behavior patterns of user behavior to understand strategies the users’ employ for learning and problem solving. Most of the work discussed above is performed offline using a variety of data mining and machine learning techniques. The patterns and users’ learning categories form the basis of designing adaptive tutors that can track user behaviors online and provide support at opportune moments to help develop better learners and problem solvers. In the framework of self-improving systems we propose two extensions: (1) after initial offline learning, move the user characterization and behavior learning algorithms online, so that the system can continue to learn new user characteristics and learning behaviors as additional learners use the system; and (2) use a combination of Monte Carlo Tree Search (MCTS) and Refinement Learning mechanisms to refine learner models, especially in situations where there is a lack of sufficient data to learn accurate models.