| This dissertation is focused on the development of system level architectures and models of dynamically self-reconfigurable systems for machine intelligence. This research is significant for building brain-like intelligent systems. Although the development of deep submicron very large scale integration (VLSI) system, nanotechnology and bioinformatics facilitate building such intelligent systems, yet it is very challenging to study how these kinds of complex, reconfigurable systems can self-develop their connectivity structures, accumulate knowledge, make associations and predictions, dynamically interact with environment, and self-control to accomplish desired tasks.; A new framework of "learning-memory-prediction" for machine intelligence is proposed in this research, and it serves as the foundation for building intelligent systems through learning in dynamic value systems, memorizing in self-organizing networks, and predicting in hierarchical structures. These systems are characterized by on-line data driven learning, distributed structure of processing components with local and sparse interconnections, dynamic reconfigurability, self-organization, and active interaction with environment.; Learning is the fundamental element for biologically intelligent systems. The proposed online value system is able to learn and dynamically estimate the value of any multi-dimensional data set, and such value system can be used in reinforcement learning. Feedback mechanism is introduced in the self-organizing learning system to allow the machine to be able to memorize information in its distributed processing elements and make associations. After the information is learned and stored in the associative memory, a biologically-inspired anticipation-based temporal sequence learning architecture is proposed. All systems proposed in this research are hardware-oriented. A novel computing paradigm that can achieve low power consumption for designing large scale, high density intelligent systems is, proposed, and a brief description of the system level hardware architecture for prototyping and testing of the proposed systems is also presented.; Intelligent systems have wide applications from military security systems to civilian daily life. In this research, different application problems, including pattern recognition, classification, image recovery, and sequence learning, are presented to show the capability of the proposed systems in learning, memory, and prediction. |
