CS ATLAS A map of computing

Domain 10 —

Artificial Intelligence

8 topics1956 — 2017
1956Dartmouth & Symbolic AI1958Perceptron & Connectionism1965Expert Systems & AI Winter1986Deep Learning1988Statistical Learning1989Reinforcement Learning2014Generative Models2017Transformers & Large Language Models

Chronology — 35 entries

Chronology

Every milestone in this domain, ordered by date. An event shared by several topics is counted once.

1943The McCulloch–Pitts neuronNeural activity described with logic gates.Perceptron & Connectionism
1956The Dartmouth workshopArtificial intelligence gets its name.Dartmouth & Symbolic AI
1957GPS, the General Problem SolverNewell and Simon.Dartmouth & Symbolic AI
1958LISPThe programming language of AI.Dartmouth & Symbolic AI
1958The perceptronA learnable linear classifier.Perceptron & Connectionism
1965DENDRALAn expert system for mass spectrometry.Expert Systems & AI Winter
1966ELIZAOne of the first chatbots.Dartmouth & Symbolic AI
1969Perceptrons (the book)It points out the limits of single-layer networks.Perceptron & Connectionism
1972MYCINRule-driven medical diagnosis.Expert Systems & AI Winter
1980XCONThe commercial peak of expert systems.Expert Systems & AI Winter
1986BackpropagationMulti-layer networks become trainable.Perceptron & Connectionism
1987The AI winterThe market for specialised hardware collapses.Expert Systems & AI Winter
1988Bayesian networksProbabilistic graphical models.Statistical Learning
1989LeNetA convolutional network reads handwritten digits.Deep Learning
1989Q-learningModel-free temporal-difference control.Reinforcement Learning
1992TD-GammonAn early success at self-play.Reinforcement Learning
1995Support vector machinesKernels and the maximum margin.Statistical Learning
2000Gradient-boosted treesThe workhorse for tabular data.Statistical Learning
2001Random forestsThe face of ensemble learning.Statistical Learning
2006Deep belief networksLayer-by-layer pretraining.Deep Learning
2012AlexNetGPU-trained deep networks begin.Deep Learning
2013DQNDeep reinforcement learning plays Atari.Reinforcement Learning
2014GANsAdversarial training generates images.Generative Models
2014VAEsVariational inference with latent variables.Generative Models
2014AttentionBahdanau applies it to machine translation.Transformers & Large Language Models
2015ResNetResidual connections train very deep networks.Deep Learning
2016AlphaGoIt beats Lee Sedol.Reinforcement Learning
2017AlphaZeroGeneral game learning from scratch.Reinforcement Learning
2017TransformerThe self-attention architecture.Transformers & Large Language Models
2018BERT and GPTPretrain then fine-tune; pretrain then generate.Transformers & Large Language Models
2020DDPMDiffusion models.Generative Models
2020GPT-3Scale brings few-shot ability.Transformers & Large Language Models
2021CLIP and DALL·EAligning text and images.Generative Models
2022Stable DiffusionOpen-source image generation.Generative Models
2022ChatGPTLarge models reach the mass market.Transformers & Large Language Models

Curriculum

Courses and labs

Which university courses this domain usually maps to, and whether the practical work is a standalone project course or an in-class lab.

In-class labs + a project

Labs are scaffolded assignments plus a project: the assignments give you the framework and the tests, the project usually reproduces a paper or enters a competition.

Courses

Introduction to AIMachine LearningDeep Learning (elective)

Open courses

  • Berkeley CS 188 Introduction to AI
  • Stanford CS 231N Deep Learning for Computer Vision

What the labs look like

  • The Pacman assignments

    Berkeley CS 188: search, games, MDPs and reinforcement learning, all inside one game

  • CS231n assignments

    From kNN to CNN and RNN, with backpropagation written by hand and no autograd allowed

  • Paper reproduction or contest

    The usual project: reproduce a conference paper or submit a Kaggle score