|
Abstract : |
Learning directly from various sensors, visual, auditory, tactile, etc., plays a central role in the development of human's intelligence. In contrast to approaches of hand-crafting knowledge-level rules or models for complex intelligent systems, the SHOSLIF approach uses a comprehensive learning approach, which characterizes signal-level representation, learning and system self-organization but minizes imposing hand-crafted knowledge-level rules or models. This paper explains why comprehensive learning is crucial for a vision system to be capable of operating in complex real-world environments; how to automatically select the most useful features from high dimensional inputs; and how to automatically organize information using a coarse-to-fine space partition tree which results in a very low, logarithmic time complexity for retrieving from a large visual knowledge base. SHOSLIF has been applied to a variety of problems. Some preliminary results are reported here for the problems of object recognition, motion understanding and autonomous navigation., |
