A novel memory-based pattern recognition system

This thesis proposes a novel method for learning and pattern recognition. The algorithm presented relies entirely on memory arranged in a custom hierarchical data structure which shifts the workload from the processor to memory. The structure and functionality draw on biology and neuroscience for inspiration while not losing sight of the inherent strengths and limitations of modern computers. A hierarchy of learned nodes is built, stored, and used for recognition without the need for complicated math or statistics. Recognition and prediction are inherent to the hierarchy and require little additional computation, even for matching of partial patterns. The experiments and results presented empirically demonstrate the robustness of memory-based recognition of images