Deep learning is machine learning built on neural networks with many layers. Rather than having engineers hand-design the features a model should look for, deep networks learn their own layered representations directly from raw data, with each layer building more abstract features on top of the last. This ability to learn features automatically is what made deep learning so powerful across vision, speech, and language.
The field’s defining overview is the 2015 paper “Deep learning” by Yann LeCun, Yoshua Bengio, and Geoffrey Hinton, published in Nature, volume 521. Written by three of the field’s founders, it explains how layered networks learn representations and surveys their breakthroughs, making it an authoritative primary statement of what deep learning is and why it works.
Deep learning is the technical core of the AI surge that began in the 2010s. It connects the older ideas of perceptrons and backpropagation to the convolutional networks, sequence models, and transformers that drive today’s products.
Why business readers should care: when companies talk about “AI” today, they almost always mean deep learning. Its hunger for data and computing power explains both the remarkable capabilities and the heavy infrastructure costs of modern AI systems.