PeopleofSuperintelligence

Researcher

Alex Graves

Also known as Alexander Graves

Alex Graves is an AI researcher whose work spans sequence learning, memory-augmented networks, speech generation, reinforcement learning and generative models. Early in his career, he studied theoretical physics at the University of Edinburgh and artificial intelligence at IDSIA, then held a postdoctoral post with Geoffrey Hinton at the University of Toronto. His research includes Connectionist Temporal Classification, Neural Turing Machines, WaveNet, Bayesian Flow Networks and protein sequence modeling. In 2026, ICML gave the Test of Time Award to his coauthored paper on asynchronous deep reinforcement learning.

Works
90
Citations
89,767
h-index
47

Publication numbers from OpenAlex, as of 14 October 2026.

Photo of Alex Graves

Timeline

  1. By 2025

    From physics into AI

    He earned a BSc in theoretical physics at the University of Edinburgh and a PhD in artificial intelligence at IDSIA under Jürgen Schmidhuber. He held a postdoctoral post with Geoffrey Hinton at the University of Toronto.

  2. 2004–2013

    Labeling sequences directly

    At ICML 2006, he coauthored Connectionist Temporal Classification, a method for training recurrent networks to label unsegmented sequences directly. His sequence-labeling algorithms were widely used for commercial speech and handwriting recognition.

  3. 2014–2016

    Giving networks memory

    His 2014 Neural Turing Machine linked neural networks to external memory and learned simple algorithms such as copying and sorting. Graves also contributed to DeepMind’s later Differentiable Neural Computer, a memory-augmented model that learned to answer questions about structured data.

  4. 2016

    Generating natural speech

    The WaveNet paper lists Graves as a coauthor and describes a neural network that generates raw audio waveforms. For text-to-speech, it reports human listeners rated the output as more natural than leading parametric and concatenative systems in English and Mandarin.

  5. 2016

    Learning with parallel actors

    He coauthored the paper Asynchronous Methods for Deep Reinforcement Learning. Its strongest actor-critic version surpassed the current Atari result while training for half the time on a single multicore CPU.

  6. 2023

    Building Bayesian Flow Networks

    In 2023, he and coauthors introduced Bayesian Flow Networks, a generative model that updates independent distributions through Bayesian inference and passes them to a neural network.

  7. 2025

    Applying generation to proteins

    By 2025, he was a senior staff researcher at InstaDeep focused on generative AI for biology; earlier, he served as chief scientist at NNAISENSE and a senior researcher at DeepMind. He coauthored ProtBFN, a 650M-parameter model trained on UniProtKB protein sequences to generate novel, structurally coherent proteins.

  8. 2026

    An award for asynchronous learning

    ICML gave its 2026 Test of Time Award to Asynchronous Methods for Deep Reinforcement Learning, listing Graves among the authors.

Sources and credits

Photo: cs.toronto.edu; Home Page of Alex Graves; creator not stated. License: Not stated by source. Source: cs.toronto.edu. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. MenaML - Speakers 2025.mena.ml. Checked 14 October 2026.
  2. Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks mlanthology.org. Checked 14 October 2026.
  3. Neural Turing Machines arxiv.org. Checked 14 October 2026.
  4. Differentiable neural computers deepmind.google. Checked 14 October 2026.
  5. WaveNet: A Generative Model for Raw Audio arxiv.org. Checked 14 October 2026.
  6. Asynchronous Methods for Deep Reinforcement Learning arxiv.org. Checked 14 October 2026.
  7. Bayesian Flow Networks arxiv.org. Checked 14 October 2026.
  8. Protein sequence modelling with Bayesian flow networks nature.com. Checked 14 October 2026.
  9. ICML 2026 Awards icml.cc. Checked 14 October 2026.