摘要
Recent years have witnessed significant developments in deep learning and artificial intelligence[1].For instance,remarkable improvements have been made in automated face comparison systems by using deep learning,compared with the classic approaches[2].The term“deep learning”is often used to refer to certain kinds of neural networks.The first publications on biological neural networks and the brain date back to the late 1800s[3].It was not until the rediscovery of the back-propagation algorithm[4]in 1986 that interest in the field was reignited.An artificial neural network is designed following a simple modelling of the brain,and involves a representation of neurons.A neuron receives a specific signal and converts to a different one.Neurons can also be used to learn from examples.They adjust a transfer function.Many neurons are linked together,and are often used in multiple layers.A visual overview is provided in Figure 1.An example of the application of neural networks is face recognition[5],where these networks examine images of people’s faces and find features,such as shapes of nose,ears,and mouth.In such networks,the parameters of thousands or more neurons are adjusted based on training to improve recognition performance.Combined with improved pattern recognition to detect the eyes,mouth,and the position of the face,they yield better results.
