# Deep Learning Deep learning is the branch of [[Machine Learning]] that uses [[Neural Network]]s with many layers ("deep" stacks of them) to learn increasingly abstract features of data. In an image network, early layers detect edges, middle layers detect shapes, and later layers detect objects — a hierarchy of representations learned automatically rather than designed by hand. Deep learning is the engine of essentially every AI breakthrough of the 2010s–2020s: image recognition, speech, machine translation, game-playing, and the large language models behind the current boom. Three things converged around 2012 to make it work after decades of disappointment: vastly more **data** (the internet), vastly more **compute** (especially the [[Graphics Processing Unit]]), and key algorithmic advances built on Backpropagation. The watershed was AlexNet (2012), a deep Convolutional Neural Network that crushed the ImageNet image-recognition competition and triggered the field's pivot to deep learning. The 2018 Turing Award to [[Geoffrey Hinton]], [[Yann LeCun]], and [[Yoshua Bengio]] — the "godfathers of deep learning" — formalized the revolution. ## Why it matters here Deep learning's defining trait — that it gets dramatically better with *more data and more compute* (the [[Scaling Laws]]) — is the single fact driving the geopolitics of AI in this vault. If capability scales with compute, then chips become strategic weapons: this is why [[Compute Sovereignty]], the [[AI Arms Race]], and [[Chip Export Controls]] all exist, and why a [[Graphics Processing Unit]] maker ([[NVIDIA]]) became one of the most valuable companies on Earth. The "bigger is better" logic also concentrates frontier AI in the few firms that can afford billion-dollar training runs ([[OpenAI]], [[Anthropic]], [[Google DeepMind]]), shaping the industry's oligopolistic structure. Deep learning is also what makes modern surveillance ([[Facial Recognition]]) and synthetic media (deepfakes, see [[Generative AI]]) actually work. ## See Also - [[Machine Learning]] — the parent field - [[Neural Network]] · Backpropagation — the architecture and training method - Convolutional Neural Network · [[Transformer]] — the breakthrough architectures - [[Scaling Laws]] — why "bigger is better" drives the compute race - [[Graphics Processing Unit]] — the hardware that made it possible - [[Neurosymbolic AI]] — the counter-thesis that scale alone will not deliver reliable reasoning