DeepMind

17 entries between January 2016 and April 2023, 4 turning points, 13 releases.

Turning points

  1. DeepMind and EMBL-EBI publish the AlphaFold protein database

    DeepMind and EMBL’s European Bioinformatics Institute opened the AlphaFold Protein Structure Database on 22 July 2021, with more than 350,000 predicted structures covering the roughly 20,000 proteins of the human proteome and 20 other organisms used in research.

  2. AlphaFold 2 nears laboratory accuracy at CASP14

    Organizers of the fourteenth Critical Assessment of protein Structure Prediction said on 30 November 2020 that DeepMind’s AlphaFold 2 had predicted about two-thirds of the competition’s target proteins to an accuracy comparable to laboratory methods.

  3. AlphaGo Zero learns Go from self-play alone

    DeepMind announced AlphaGo Zero, trained only by playing itself from random play, with no human games and no hand-crafted features. After three days it beat the version that had defeated Lee Sedol, 100 games to nil. Nature published the work the next day.

  4. AlphaGo defeats Lee Sedol 4-1 in Seoul

    DeepMind’s AlphaGo won the last game of a five-game match against Lee Sedol in Seoul on 15 March 2016, taking the series 4-1. Lee, a nine-dan professional, won game four, the only game the program lost. The winner’s $1 million prize went to charity.

Every entry

  1. Google merges Brain and DeepMind into Google DeepMind

    Google said on 20 April 2023 that it was combining the Brain team from Google Research with DeepMind into a single unit called Google DeepMind, led by DeepMind’s co-founder and chief executive Demis Hassabis. Google said the move would accelerate its progress in AI.

  2. AlphaFold database expands to over 200 million structures

    DeepMind and EMBL’s European Bioinformatics Institute expanded the AlphaFold Protein Structure Database on 28 July 2022, from roughly 1 million predicted protein structures to more than 200 million, covering nearly every sequence catalogued in UniProt.

  3. Chinchilla finds most large language models are undertrained

    DeepMind posted “Training Compute-Optimal Large Language Models” on 29 March 2022. Its 70-billion-parameter Chinchilla, trained on four times the data used for the 280-billion-parameter Gopher at equal compute cost, outperformed Gopher and GPT-3 across a broad set of benchmarks.

  4. DeepMind and EMBL-EBI publish the AlphaFold protein database

    DeepMind and EMBL’s European Bioinformatics Institute opened the AlphaFold Protein Structure Database on 22 July 2021, with more than 350,000 predicted structures covering the roughly 20,000 proteins of the human proteome and 20 other organisms used in research.

  5. DeepMind’s MuZero plays Atari, Go, chess and shogi without rules

    Nature published “Mastering Atari, Go, chess and shogi by planning with a learned model” on 23 December 2020. Julian Schrittwieser and colleagues at DeepMind described MuZero, which planned using a model it learned rather than rules it was given.

  6. AlphaFold 2 nears laboratory accuracy at CASP14

    Organizers of the fourteenth Critical Assessment of protein Structure Prediction said on 30 November 2020 that DeepMind’s AlphaFold 2 had predicted about two-thirds of the competition’s target proteins to an accuracy comparable to laboratory methods.

  7. DeepMind releases AlphaFold predictions for coronavirus proteins

    DeepMind used AlphaFold to predict the shapes of several under-studied SARS-CoV-2 proteins, among them the membrane protein, Nsp2, Nsp4, Nsp6 and the papain-like proteinase, and released them under an open license as computational predictions rather than verified structures.

  8. AlphaStar defeats professional StarCraft II players

    DeepMind said on 24 January 2019 that its AlphaStar system had beaten the professional StarCraft II players Dario Wünsch and Grzegorz Komincz, winning 10 of 11 games played under professional conditions in December 2018. It was trained by league play among versions of itself.

  9. AlphaFold wins the CASP13 protein-structure contest

    DeepMind’s AlphaFold placed first among entrants at the thirteenth Critical Assessment of protein Structure Prediction in December 2018, a blind test of methods for predicting how a protein folds. It did best on the hardest category, targets with no similar known structure.

  10. AlphaZero learns chess and shogi from the rules alone

    DeepMind posted a paper describing AlphaZero, one algorithm that, given only the rules, reached superhuman play at chess, shogi and Go within 24 hours of self-play training and beat the reigning computer champion program in each game.

  11. AlphaGo Zero learns Go from self-play alone

    DeepMind announced AlphaGo Zero, trained only by playing itself from random play, with no human games and no hand-crafted features. After three days it beat the version that had defeated Lee Sedol, 100 games to nil. Nature published the work the next day.

  12. UK regulator finds Royal Free’s DeepMind data deal unlawful

    The Information Commissioner’s Office ruled on 3 July 2017 that the Royal Free London NHS Foundation Trust broke the Data Protection Act by giving DeepMind identifiable records for about 1.6 million patients to test Streams, a kidney-injury app. DeepMind accepted the finding.

  13. AlphaGo defeats Ke Jie 3-0 in Wuzhen

    AlphaGo played Ke Jie, then the world’s top-ranked Go player, in three games at the Future of Go Summit in Wuzhen, China, taking the first by half a point and the next two by resignation. DeepMind had offered a $1.5 million prize.

  14. Five companies found the Partnership on AI

    Amazon, Facebook, IBM, Microsoft and Google’s DeepMind announced the Partnership on Artificial Intelligence to Benefit People and Society, a non-profit for best practice and public understanding, open to academic and civil-society members. Apple joined in January 2017.

  15. DeepMind publishes WaveNet, a model of raw audio

    DeepMind published WaveNet, a generative network that modeled raw audio waveforms directly. DeepMind said it closed more than half the quality gap between existing text-to-speech systems and recorded human speech, and that it also generated music.

  16. AlphaGo defeats Lee Sedol 4-1 in Seoul

    DeepMind’s AlphaGo won the last game of a five-game match against Lee Sedol in Seoul on 15 March 2016, taking the series 4-1. Lee, a nine-dan professional, won game four, the only game the program lost. The winner’s $1 million prize went to charity.

  17. Nature publishes the paper behind AlphaGo

    Nature published “Mastering the game of Go with deep neural networks and tree search” by David Silver, Demis Hassabis and 18 co-authors at Google DeepMind. It disclosed that AlphaGo had already beaten the European champion Fan Hui five games to nil in October 2015.

Releases

Date Release Maker Family Kind What it was
2022
AlphaFold Protein Structure Database (200M+ update) DeepMind, EMBL-EBI dataset, open weights Expanded from about 1 million to over 200 million predicted structures
Gato DeepMind paper Single 1.2B-parameter transformer performing over 600 tasks across modalities
Chinchilla (“Training Compute-Optimal Large Language Models”) DeepMind paper 70B-parameter model shows compute-optimal scaling beats larger undertrained models
2021
AlphaFold Protein Structure Database DeepMind, EMBL-EBI dataset, open weights More than 350,000 predicted structures covering the human proteome and 20 other organisms.
2020
MuZero DeepMind paper Model-based reinforcement learning across Go, chess, shogi and Atari without being given the rules.
AlphaFold 2 DeepMind model Near-experimental accuracy on two-thirds of the CASP14 target proteins. Weights and code not yet public.
AlphaFold SARS-CoV-2 protein predictions DeepMind dataset, open weights Predicted structures for six under-studied coronavirus proteins, released openly as computational predictions.
2019
AlphaStar (Grandmaster) DeepMind paper Nature paper reporting Grandmaster level across all three StarCraft II races under professional conditions.
AlphaStar DeepMind model Beat the professional StarCraft II players MaNa and TLO 10-1 in matches played the previous December.
2017
AlphaZero DeepMind paper One self-play algorithm reaching superhuman play at chess, shogi and Go from the rules alone.
AlphaGo Zero DeepMind paper A version of AlphaGo trained from self-play alone, with no human game records as input. Published in Nature.
2016
WaveNet DeepMind paper A generative model of raw audio waveforms, used for speech synthesis and for music.
AlphaGo (Nature paper) DeepMind paper Deep neural networks paired with tree search. The method behind the Lee Sedol match, and the first disclosure of the 5-0 win over Fan Hui.