Convolutional neural network

A neural network that applies learned filters across spatially organized data. Shared weights let it detect local patterns, such as edges, wherever they occur in an image.

Convolution: A mathematical operation that combines a function with a shifted, weighted copy of another function. The network’s defining operation applies learned filters across input locations.
Artificial neural network: A computational model of interconnected units whose adjustable parameters transform inputs into outputs. Convolutional networks are a specialized architecture within this broader model family.
Image classification: The task of assigning one or more semantic categories to an image. Convolutional networks became widely used for learning image categories from pixels.
Fully connected neural network: A neural network layer in which every input unit connects to every output unit. Unlike convolutional layers, it does not share weights across spatial locations.
Receptive field (neurophysiology): The region of sensory input that changes the activity of a neuron or sensory cell. Early visual neuroscience described local response regions that inspired convolutional design.
Convolution kernel: A small array of weights applied across an input to compute local responses. Each learned kernel responds to a particular local pattern.
Tensor: A multidimensional array used to represent numerical data and parameters in computing. Images, feature maps, and filter weights are represented as tensors.
Object detection: The task of locating and identifying objects within an image or video frame. Convolutional features support both object localization and category prediction.
Recurrent neural network: A neural network architecture that processes sequences using recurrent connections or state. Recurrent networks model sequence order directly rather than relying on spatial filters.
Hubel and Wiesel: David Hubel and Torsten Wiesel were neuroscientists who studied visual processing in the cat cortex. Their findings on receptive fields and orientation-selective cells influenced early visual models.

Linked from 10 pages