Pattern Recognition Models for Bioinspired Computing and Document Analysis

点击翻译按钮获取中文摘要

We propose a novel form of bioinspired computing that models the structural adaptability of biological systems. This research addresses the problem that the network properties of fascia (connective tissue) are poorly understood. Our computational model of the structural responsiveness of a biological system leads to advances within computer science, and also contributes insights relevant to applications in medicine, physiology and biological modeling.**Our novel form of bioinspired computing is called Fascial Network Computing by analogy with Neural Network Computing, a well-established form of bioinspired computing that models the neural adaptability of biological systems. Fascia is a bodywide network of connective tissue that provides structural support, protection, shock absorption and elastic recoil. Fascial tissue adapts by changing its characteristics in response to the demands placed on it. This inspires our training algorithm for a simulated fascial network: sections of fascia that are frequently under high load respond by increasing their stiffness. Analogously, during training of a neural network, neurons that are frequently co-activated respond by increasing the strength of the connection between them.**The pattern of stiffness in a trained fascial network constitutes a type of distributed memory, analogous to the distributed memory formed by the pattern of connection strengths in a trained neural network. Both neural and fascial networks exhibit emergent properties such as global response to local injury. Fascial Network Computing offers insight into the structural responsiveness of a biological system, an intriguing complement to the neural responsiveness modeled by Neural Network Computing. **The first long-term goal is to define Fascial Network Computing as adaptive tensegrity, and investigate its computational properties to advance the state of the art in bioinspired computing. The second long-term goal is to extend Fascial Network Computing to include an abstract model of injury, thereby producing knowledge that increases the accuracy of computer-based concussion models. **The requested funding will support the training of two PhD, five MSc and four undergraduate students in leading edge bioinspired computing research. I will train students in technical skills by drawing on my wide range of experience with topics such as pattern recognition, document analysis, abstraction in modeling, model validation, classifier combination, biomedical document retrieval and holographic reduced representations. In addition, by placing emphasis on fostering enthusiasm, confidence, discipline, information acquisition and technical communication in my students, I develop a well-balanced foundation for their success in research.