2 - Understanding artificial neural networks: analogy to the biological neuron model

Abstract

Brief concepts of neural networks are introduced for engineering applications by which a real-world designing of diverse structures and frames is performed. By studying materials presented in this chapter, readers will grasp at AI technologies including big data generation, training, feedforward networks, and backpropagation required to understand for deriving artificial neural networks (ANNs) with multiple layers and neurons. Readers will learn what ANNs are, how ANNs are trained, and how ANNs are tested and validated. Important concepts including neurons, weights, biases, and activation functions should be understood when deriving ANNs. Large structural datasets are initialized and normalized before being used to train ANNs. Finally, feedforward networks are trained and corrected by backpropagation to obtain accurate weight and bias matrices for engineering applications. Many examples are presented to guide readers with engineering designs using ANNs. Good opportunities for undergraduate and graduate students to understand AI-based data-centric engineering to face challenging issues.