# The Deep-Learning Revolution (2006–2020)

Deep neural networks, accelerated by GPUs and large datasets, transformed AI from 2006 onward. This sub-article expands [Section 6 of History](../history.md).

## The Breakthrough
- Hinton's 2006 work on **deep belief networks** showed deep networks could be trained effectively.
- **AlexNet** (2012) crushed prior benchmarks on ImageNet, igniting the field.

## Hardware Acceleration
- **GPUs** proved ideal for matrix-heavy neural computation; CUDA made them programmable.
- Large clusters and specialized accelerators (see [Specialized AI Hardware](specialized-ai-hardware.md)) scaled training.

## Frameworks
- Theano, Caffe, **TensorFlow** (2015), and **PyTorch** (2016) democratized model building.

## Domain Successes
- **CNNs** transformed image recognition and segmentation.
- **RNNs/LSTMs** improved sequence modeling; then **Transformers** (2017) revolutionized NLP and beyond. See [Transformers](transformers.md).
- **AlphaGo** (2016) combined deep learning and reinforcement learning to defeat Lee Sedol.

## Why It Matters
Deep learning replaced hand-crafted features with learned hierarchies of representation, and scale became the primary driver of progress.

## Sources
- Krizhevsky, A. et al. (2012). *ImageNet Classification with Deep Convolutional Neural Networks.*
- Hinton, G. et al. (2006). *A fast learning algorithm for deep belief nets.*
- Vaswani, A. et al. (2017). *Attention Is All You Need.*

*Back to [History](../history.md) · Next: [Generative & Agentic Era](generative-agentic-era.md).*