Harri Valpola, ZenRobotics Ltd.
From neural networks to artificial intelligence
Symbolic artificial intelligence and neural networks research have both aimed at intelligent systems but their starting points have been very different. Symbolic AI has focused on high-level cognition, symbol manipulation, reasoning and planning, while neural networks research has studied perception and learning. In this talk I will discuss a new generation of neural networks which are approaching higher levels of cognition. They are deeply rooted in neural networks tradition: parallel distributed processing, analog representations and focus on learning from real-world data. However, unlike most current neural networks, the new generation of networks relies on emergent dynamic phenomena which correspond to symbols, role binding and manipulation of structured representations. In other words, we are again tackling the same problems as symbolic AI but this time we are firmly grounded in real-world data and learning.
Academy research fellow Harri Valpola, PhD, is the leader of computal neuroscience group in Aalto University School of Science and Technology. The group studies the information processing principles of the brain and applies them in robotics. Dr. Valpola is also the chief scientist at ZenRobotics Ltd. which applies the research in intelligent, learning robots.