I am a CNRS researcher at the Ecole Normale Supérieure currently detached to Meta AI, where I lead the Brain & AI group. We work on identifying the brain and computational bases of human intelligence, with a focus on language. We develop deep learning techniques to decode and model intracranial recordings, magneto-encephalography and functional magnetic resonance imaging.


Decoding speech from non-invasive brain recordings

Défossez, Caucheteux, Kabeli, Rapin & King, arXiv 2022

Toward a realistic model of speech processing in the brain with self-supervised learning

Millet*, Caucheteux*, Orhan, Boubenec, Gramfort, Dunbar, Pallier & King, arXiv 2022

Brains and algorithms partially converge in natural language processing

Caucheteux, King, Nature Communications Biology 2022

Long-range and hierarchical language predictions in brains and algorithms

Caucheteux, Gramfort & King, arXiv 2021

tl;dr:We track language predictions in the brain and show that, unlike modern algorithms, they are hierarchical and apply to a variety of temporal scopes.

Model-based analysis of brain activity reveals the hierarchy of language

Caucheteux, Gramfort & King, EMNLP 2021

tl;dr:We show how deep language algorithms help reveal the hierarchical organization of language integration in the brain.

GPT-2’s activations predict the degree of semantic comprehension in the human brain

Caucheteux, Gramfort & King, bioRxiv 2021

tl;dr:The more we understand text, the more our brain responds like GPT-2.

Disentangling Syntax and Semantics in the Brain with Deep Networks

Caucheteux, Gramfort & King, ICML 2021

tl;dr:The similarity between deep nets and the brain allow us to decompose syntax and semantics in the brain.

Inductive biases, pretraining and fine-tuning jointly account for brain responses to speech

Millet & King, arXiv 2021

tl;dr:Do convolutional networks process speech sounds like our brains does? Short answer: yes, even without training; but training helps.

Deep Recurrent Encoder: A scalable end-to-end network to model brain signals

Chehab*, Defossez*, Loiseau, Gramfort & King, arXiv 2021

tl;dr: We propose a new end-to-end architecture to encode MEG brain signals. It outperforms standard pipelines by a 3X.

Bifurcation in brain dynamics reveals a signature of conscious processing independent of report

Sergent, Corazzol, Labouret, Stockart, Wexler,King, Meyniel & Pressnitzer , Nature Communications 2021

tl;dr: We show with EEG that the conscious access follows an all-or-none dynamics even without report.

Language processing in brains and deep neural networks: computational convergence and its limits

Caucheteux & King, bioRxiv 2020

tl;dr: Do deep nets become increasingly correlated with brain activity as they learn to process language? Short answer: only their middle layers do.

Back-to-back regression: Disentangling the influence of correlated factors from multivariate observations

King, Charton, Lopez-Paz & Oquab, Neuroimage 2020

tl;dr: We introduce a simple method to combine the advantages of decoding and encoding analyses.

COVID-19: the promises and pitfalls of Machine Learning

Peiffer-Smadja, Maatoug, Lescure, D’Ortenzio, Pineau & King, Nature Machine Intelligence 2020

tl;dr: We’re teaming up with the AP-HP hospital to review the promises and pitfalls of Machine Learning.

Neural dynamics of phoneme sequencing

Gwilliams, King, Marantz & Poeppel, bioRxiv 2020

tl;dr: Decoding the neural dynamics underlying phonetic representations shows how the brain can keep up multiple phonemes until the corresponding word is identified.

Intersecting AI and Neuroscience

The Human Brain Encodes a Chronicle of Visual Events at Each Instant of Time Through the Multiplexing of Traveling Waves

Wyart and King, Journal of Neuroscience 2021

tl;dr: We measure brain responses to image sequences, and show how the brain recruits a hierarchy of neural processes in order to efficiently represents multiple snapshots of the past. Check-out our tweet thread for the illustrated summary

Recurrent Processes Emulate a Cascade of Hierarchical Decisions

Gwilliams and King bioRxiv 2019

tl;dr: When an image is ambiguous, the brain slowly recruits a hierarchy of recurrent processes to generate categorical percepts. Check-out our tweet thread for the illustrated summary

Detection of Brain Activation in Unresponsive Patients with Acute Brain Injury

Claassen et al, New England Journal of Medicine 2019

tl;dr: Acute brain injury patients can sometimes be behaviorally unresponsive. Yet, we show that 15% of them still demonstrate motor-command brain responses.