Faculty of Engineering and Natural Sciences
Automatically Creating Knowledge-Rich AI Systems Using Relational Machine Learning
Tolga O. Könik
Center for the Study of Language and Information, Stanford University
Developing AI systems that function autonomously and intelligently in complex environments is a difficult process that requires substantial programming expertise and development time. In this talk, I present a machine learning framework that automates this process. We expect that the target AI systems are capable of processing complex knowledge structures, which most mainstream machine learning techniques fail to represent properly. Fortunately, recent machine learning research usually treated under the umbrella headings such as relational learning or inductive logic programming (ILP) provides a framework that can address this issue. ILP combines complex knowledge structures, logical reasoning and machine learning using a first order logical representation.
In this talk, I present a framework for learning by observation that automatically creates AI programs from behaviors of experts. I describe how an ILP approach that I call “relational learning by observation” addresses some of the most critical challenges of learning by observation problem such as that the expert’s mental reasoning is not directly available to the learner.
I also describe another framework for AI program creation, where a human expert specifies abstract scenarios describing desired behavior of the target AI system using a diagrammatic storyboard-like representation. This approach is an example of a new paradigm for programming AI systems where the expert/programmer transfers his/her knowledge to the AI system using a graphical interface. Although this approach uses the same underlying learning system with our first approach, its graphical interface provides new mechanisms for the expert to communicate his/her reasoning. Moreover, it is a more interactive approach where a previously learned agent program helps the expert in specifying the scenarios. The agent program gives immediate feedback on how it would react to the specified situations, helping the expert to generate more relevant behavior data.
Tolga O. Könik graduated from Bogazici University with a B.S. degree in Electrical Engineering and a B.S. degree in Mathematics in 1997. He earned his M.S. degree in Systems and Control Engineering from the same institution. In 2001, he received an M.S. degree in Computer Science from University of Michigan. He is expected to receive his Ph.D. degree in Computer Science from the same institution in July 2006. His dissertation work involves learning by observation in cognitive agent. Since August 2005, he is working as a Research Scientists at the Center for the Study of Language and Information in Stanford University. He is also associated with the Institute for the Study of Learning and Expertise. His most recent work involves studying how knowledge can be transferred between different machine learning tasks. His current research interests include Machine Learning, Relational Machine Learning, Inductive Logic Programming, General Cognitive Agent Architectures, Learning in Cognitive Agents, Common Sense Reasoning, and Qualitative Reasoning.
April 26, 2006, 13:40, FENS 2019