Abstract
This chapter presents the fundamentals of a hardware based memory network that can perform complex cognitive tasks. The network is designed to provide space dimensionality reduction, which enables desired functionality in a random environment. Complex network functionality is achieved by simple network cells that minimize the needed chip area for hardware implementation. Functionality of this network is demonstrated by automatic character recognition with various input deformations. In the character recognition, the network is trained to recognize characters deformed by random noise, rotation, scaling, and shifting. This example demonstrates how cognitive functionality of a hardware network can be achieved through an evolutionary process, as distinct from design based on mathematical formalism.
TopComplex networks such as neural networks and cognitive memory networks can be classified as a network based approach to information processing, often finding its way in formulating inter-relationships within an intelligent systems framework. The ability of making decisions is the key aspect in understanding the interpretation drawn from the network analysis. Clearly, information as we know today is a concept that is perceived to be useful to make informed judgments at local and global decision levels. The hierarchical processing of information is a unique and interesting aspect of the human brain, which is by at large implemented in the layered structure of the cortex. Each module in a human brain has evolved in structure and is trained to process and understand the information specific to a given task. The interaction between the modules results in the ability of the human brain to understand and respond to the complex cognitive task. The interaction of the module follows a hierarchical organization, where sensory processes can be visualized as the bottom part of the hierarchy and decision making processes can be visualized as the top most part of the hierarchy. This talk provides insights into the new paradigm of thinking in machine based information processing methods. In this chapter, we present the very ideas of human intelligence in perspective of semiconductor memory elements that can be used for creating real-time intelligent information procession units and systems.
Key Terms in this Chapter
Character Recognition: Automatic techniques of matching and recognition hand written and printed natural language characters.
Machine Intelligence: A multidisciplinary subject of study with a focus on creating and enabling machines with human like intelligence.
Neural Network: A network of neurons or a hardware or mathematical model that represents such a network.
Neuromorphic Circuit: Circuits that mimic the functionality and electrical properties of a biological neuron.
Pattern Matching: Science and technique in calculating the similarity and establishing the relation between two patterns.
Memory Network: A network consisting of semiconductor memories that has the properties of modularity and hierarchy at structural and functional levels.