Chapter 3 – Adapting Biological Principles for Deployment in Computational Science

Embedding biological principles into mathematical methods has yielded valuable tools for scientists and engineers. Neural nets mimic human learning processes. They are mathematical constructs that simulate natural adaptive learning. Nets are trained iteratively on input data along with the corresponding target outcomes. After a sufficient number of training iterations, nets learn to recognize patterns (if they exist) in the data and, in effect create, internal models of the processes governing the data. Trained nets can be used to predict outcomes for fresh input conditions, just as the child eventually learns to recognize shapes in general. The concepts for developing adaptive learning systems are derived from the understanding of biological neural cognitive processes. To obtain a primitive view of how biological neural systems process signals, imagine an ensemble of neurons connected to each other. Adaptive learning algorithms consist of the following building blocks: simulated neurons, connections among these neurons, and the weights that modify the signals passed through these connections. During the learning process, the weights are modified so that the net converges on learning the desired behavior. Hence, the “information content” of the net is embodied in these weights.