Integrated learning systems that anticipate

Machines are often inflexible since they can not learn. My research goal is to understand and to build learning systems. Learning systems that we want to understand better includes biological organisms, in particular humans, where learning is an essential part of cognitive functions, and also machines, where learning can provide flexible solutions to new problems. The more specific research goal outlined in this proposal is to build learning system that can better guide decisions in complex environments. The system will integrate three major factors of learning, learning about sequences instead of static information more common in current systems, learning hierarchical representations of the environment to provide an efficient way of using such information, and combine such hierarchical temporal memories with learning that is guided by reward. The system will be applied to autonomous robots with the aim to search, identify and map objects in unknown territories. An example application is the identification of artificial objects under water from sonar data gathered by autonomous underwater vehicles (AUVs).There has been many new algorithms to learn from the environment in various circumstances. This research aims to integrate systems that are flexible enough to work in real world environments, such as household robots or robots in the health care system rather than manufacturing applications that are typically more controlled. The hybrid systems proposed here are aimed at learning about the environment to find novel solutions that have not been programed explicitly into the robots. This research program has the potential to advance our understanding of complex learning systems that operate in natural environments.The research is based on a combination of statistical methods, algorithmic intelligence, robotics applications and understanding of biological systems. Learning methods are widely applicable and are and excellent area for educating HQP