ARTIFICIAL INTELLIGENCE IN ARCHITECTURE AND BUILT ENVIRONMENT DEVELOPMENT 2024: A CRITICAL REVIEW AND OUTLOOK, 3rd part: Artificial neural networks
An artificial neural network is a collection of connected units or nodes called artificial neurons designed to model loosely how the neurons in a biological brain have been supposed to look and work. Like synapses in a biological brain, each connection can transmit a signal to other neurons. A deep neural network is an artificial neural network with multiple layers between the input and output layers; in a shortcut, a deep neural network makes machine learning deep learning [59, 60]. In essence, two computing principles apply in artificial neural networks today: feedforward computing and backpropagation. The goal is always to train the models generated to cope with the criteria typically inserted by vast sample datasets. Feedforward computing refers to a type of workflow without feedback connections that would form closed loops; the latter term marks a way of computing the partial derivatives during training. When training a model in the feedforward manner, the input “flows” forward through the network layers from the input to the output. By backpropagation, the model parameters update in the opposite direction: from the output layer to the input one. Backpropagation, a strategy to compute the gradient in a neural network, is a general technique; it is not restricted to feedforward networks, it works for recurrent neural networks (to be introduced soon) as well [61].