Peter van Lith

A Knowledge Engineering approach to Deep Learning

Most Deep Learning systems are based on finding regularities or behavior patterns in large bases of examples. The learned patterns are generally not semantically grounded, making meaningful communication with human users very difficult. Using RoboCup Soccer playing robots as an example, the knowledge of experienced human coaches helps in teaching the learning system to use the same symbols as employed by human coaches. This work concentrates on understanding of team- and individual robot behaviors, by creating models, that are based on real-world symbols like formations and roles. Such models serve as grounding symbols for a Deep Learning system, in which the strengths of the symbolic and sub-symbolic systems are combined.

Peter van Lith started working in ICT in 1964. He built his own computer in 1968 and became interested in Artificial Intelligence and developed some AI systems using muLisp. In 1980 he started his own AI company and was the first one to build an Expert System in the Netherlands. He developed the Lisp based ES development environment Acquaint and created over 15 Expert Systems. He developed a learning line for robotics in lower and middle grade schools and founded RoboCup Junior in the Netherlands. Currently he is a member of the RoboCup soccer team of the University of Technology in Eindhoven, where he develops deep learning software. His research focuses on modeling the behavior of competing teams on a symbolic level, combining knowledge engineering with deep learning.

Buttonlinkedin
This website uses cookies to ensure you get the best experience. Learn more