Olivier J. Hénaff

Research Scientist

DeepMind, London, UK

Despite recent progress in artificial intelligence, humans and animals still vastly surpass machine observers in their ability to reason about their environment. In particular, biological systems generalize to new concepts much more efficiently, flexibly, and robustly than their artificial counterparts. What is it about biological perception that supports these fundamental qualities?

My research aims at addressing this question by investigating distinctive properties of biological vision, and building artificial systems that capture them. Specifically, by understanding how neural representations of the environment are structured, we can assess whether current artificial systems are similarly organized, and if not, incorporate this structure into their learning paradigm.

For example, we found that human perceptual representations were structured to facilitate the prediction of natural videos, a property lacking from current artificial recognition systems. In turn, enforcing artificial representations to be more predictable enabled them to learn about new image classes much more efficiently.



Efficient Visual Pretraining with Contrastive Detection

Olivier J. Hénaff, Skanda Koppula, Jean-Baptiste Alayrac, Aaron van den Oord, Oriol Vinyals, João Carreira

International Conference on Computer Vision (ICCV), October 2021 (Oral)

Divide and Contrast: Self-supervised Learning from Uncurated Data

Yonglong Tian, Olivier J. Hénaff, Aaron van den Oord

International Conference on Computer Vision (ICCV), October 2021

Data-Efficient Image Recognition with Contrastive Predictive Coding

Olivier J. Hénaff, Aravind Srinivas, Jeffrey De Fauw, Ali Razavi, Carl Doersch, S. M. Ali Eslami, Aaron van den Oord

International Conference on Machine Learning (ICML), July 2020

Are we done with ImageNet?

Lucas Beyer*, Olivier J. Hénaff*, Alexander Kolesnikov*, Xiaohua Zhai*, Aäron van den Oord*

Tech report, June 2020. *equal contribution

Representation of visual uncertainty through neural gain variability

Olivier J. Hénaff, Zoe M. Boundy-Singer, Kristof Meding, Corey M. Ziemba, Robbe L. T. Goris

Nature Communications, May 2020

Perceptual straightening of natural videos

Olivier J. Hénaff, Robbe L. T. Goris, Eero P. Simoncelli

Nature Neuroscience, April 2019

Geodesics of learned representations

Olivier J. Hénaff, Eero P. Simoncelli

International Conference on Learning Representations (ICLR), May 2016

The local low-dimensionality of natural images

Olivier J. Hénaff, Johannes Ballé, Neil C. Rabinowitz, Eero P. Simoncelli

International Conference on Learning Representations (ICLR), May 2015 (Oral)

More info

Google Scholar