Artificial intelligence · 2012
Deep learning (AlexNet)
AlexNet cuts ImageNet error to 15.3%, and deep learning takes over AI.
Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton trained an eight-layer convolutional network on two GTX 580 GPUs with ReLU and dropout. Its 15.3% top-5 error beat the runner-up's 26.2%. Within three years nearly all of vision, speech and translation moved to deep neural networks.
What it built on 4
What it led to 21
- Deep Q-network (Atari) 2013
- Generative adversarial network 2014
- Knowledge distillation 2015
- ResNet 2015
- Tensor Processing Unit 2015
- TensorFlow 2015
- Algorithmic short video (TikTok) 2016
- Tensor Cores (Volta) 2017
- AlphaFold 2 2020
- Driverless robotaxi service 2020
- Driverless robotaxi service 2020
- Autonomous tractor 2022
- Tesla Optimus 2022
- AI materials discovery (GNoME) 2023
- Digit in commercial logistics 2024
- Electric Atlas 2024
- Neuralink first implant 2024
- AI-driven materials discovery ~2030
- Driverless taxis in most big cities ~2030
- Brain-computer interfaces restore speech ~2032
- Whole-brain emulation ~2060
Related measurements
The dashed line marks when it first worked.
Training compute of notable AI models Source: Epoch AI, Notable AI Models (CC BY 4.0)
ML hardware FLOP/s per dollar Source: Epoch AI, ML hardware (CC BY 4.0)