In this fourth workshop, we continue to explore an expanding field of research that attempts to build systems that can learn without human intervention and with little or no hard-wired domain knowledge, as would a newborn child or animal.
As we try to build increasingly complex A.I. systems, including robots and other complex cyber-physical systems (CPS), and as these systems increasingly depend upon machine learning to be effective in their intended environments, human intervention is becoming a limiting factor. In this light, the need for and potential benefits of Self-Supervised Learning (SSL) has become apparent.
How do intelligent systems, such as humans and software agents, learn about the world? They should primarily learn about themselves and their environment and how the two interact effectively by exploration and self-motivated experimentation. We continue to ignore the fundamental mechanisms underlying self-motivated learning through experience in the open world that could ultimately lead to intelligent behaviors such as (for example) how to recognize objects through visual and other senses, how to move another object to reach an object otherwise out of reach, and how to use those objects as tools, by experimenting. Higher-level learning (e.g.) the use of specialized tools, can be taught both by example and by building upon existing learned competencies. Historically, this approach has seemed difficult due to the run-time computation burden and the need for large memories. In addition, training time has been seen as limiting. Today, processors are more powerful and cheaper, and large memories are increasingly less of an obstacle.
Organizing Committee
Dr. Milan Minsky (Leela.ai)
Professor Kristinn Thorisson (Reykjavik University)
Professor Olivier Georgeon (Catholic University of Lyon, UCLy)
Dr. Paul Robertson (DOLL Inc.)
Dr. Cyrus Shaoul (Leela.ai)
Henry Minsky (Leela.ai)
Dr. Robert Laddaga (DOLL Inc.)
Call for Papers coming soon
