A deep learning neural network algorithm that uses the concept of “convolutions” to provide greater ability for machine learning models to distinguish components of a dataset, such as portions of an image. Specifically, a convolutional neural network has dozens of layers, with a special kind of layer that performs a mathematical operation called a “convolution” that is able to pick up patterns in an image such as lines, gradients, circles, or larger components such as eyes, faces, ears, and other parts. Special purpose neural network layers such as pooling layers and fully-connected layers as well as additional convolutional layers are stacked to achieve greater detail and precision. CNNs revolutionized computer vision because they need very little preprocessing or feature engineering to work and can identify key features directly on a raw image. Inspiration for the CNN concept comes from the human visual cortex in which biological neurons operate in layers to identify incrementally complex features. In addition to computer vision applications in image and video, CNNs have been applied to recommender systems, classification, natural language processing, time-series data, and brain–computer interfaces. Famous implementations of CNNs include LeNet, which Yann LeCunn developed in 1997 for handwriting recognition among other applications, and AlexNet in 2012 which broke through a key performance barrier in winning ImageNet.
Login to register for events. Don’t have an account? Just register for an event and an account will be created for you!