Deep learning is a state-of-the-art subtype of artificial intelligence and machine learning approaches that utilizes multiple layers to extract features from raw input. Deep learning includes artificial neural networks, convolutional neural networks, and deep neural networks and has a wide variety of uses such as image recognition, natural language processing, speech recognition, computer vision and object detection.
Deep learning algorithms “learn” by analyzing large amounts of data and extracting conclusions. This is done through the transformation of data into more abstract representations as the data progresses through each layer of the algorithm. In deep learning models, the “deep” refers to the number of layers that data must pass through. For example, as part of image recognition, a deep learning algorithm might try to identify stop signs through ingestion of images with and without these signs. Then, in a first layer it may abstract the individual pixels and create edges, the next layer might compose and encode arrangements of edges, the third layer may encode a red octagon, and the fourth might recognize that the selected image contains a stop sign.
By building deep learning frameworks, users can input limited amounts of data around desired outcomes, and the learning neural networks can then decode which features are most optimal for the task on its own. While there is always the ability for some fine-tuning completed by experts in the field, this allows for high performance algorithms to mostly run on their own with little oversight. By training deep learning algorithms, organizations can then transfer learning from one method to another, meaning that if an algorithm has successfully decoded what is a stop sign, they can apply that knowledge to helping identify other images, by recognizing that it is not a stop sign.
Deep learning has a variety of uses, including: