A type of deep learning model that is specifically designed for processing data that has a grid-like structure, such as images. Inspired by the human brain and dating back to the mid 20th century, it is simply a collection of nodes connected together with a combination of linear algebra, non-linear activations, and model weights. A simple example would be a model that predicts whether someone should play tennis today. The input nodes could be today’s temperature, the chance of rain, distance to the nearest court, and the person playing. The interior math and model weights would just be learned calculations about how important temperature is compared to rain, and how the changes of those two interact to create the final likelihood. Modern neural networks follow these same ideas, they have just increased in size and in cleverness of the assumptions baked into the architectures.
