Welcome!
I am a postdoc working at the interface of neuroscience and artificial intelligence, at Facebook AI in New York.
*I am starting a lab soon at Aalto University in Helsinki. Don't hesitate to reach out for more information!*
On my mind currently: how to build neural networks for vision which would work as well as our visual system...
Brief CV (full CV)
Now - Postdoc at Facebook AI in New York City, advised by Yann LeCun.
2019 - Postdoc in the department of Applied Physics at Stanford University, advised by Surya Ganguli.
2017 - PhD in computational neuroscience at Paris VI University and the Vision Institute, advised by Serge Picaud and Olivier Marre.
2012 - MS in computer science from Supelec (France).
2009 - BS in mathematics and physics (MPSI/MP) from Lycée Hoche, Versailles (France).
Featured Publications
Barlow Twins: Self-Supervised Learning via Redundancy Reduction
Jure Zbontar*, Li Jing*, Ishan Misra, Yann LeCun, Stéphane Deny
A simple method for state-of-the-art self-supervised learning, inspired by a 60-year-old principle from neuroscience. more on Twitter...
Addressing the Topological Defects of Disentanglement via Distributed Operators
Diane Bouchacourt*, Mark Ibrahim*, Stéphane Deny
arXiv 2021 / Code / PDF
We use topological arguments to show that disentanglement as commonly defined introduces discontinuities in the encoder, which leads us to propose a new approach to disentanglement through distributed equivariant operators. more on Twitter...
A Unified Theory of Early Visual Representations from Retina to Cortex through Anatomically Constrained Deep CNNs
Jack Lindsey*, Samuel A. Ocko*, Surya Ganguli, Stéphane Deny
International Conference on Learning Representations (ICLR 2019) / Code / PDF
Visual representations differ drastically between the retina and primary visual cortex, and retinal representations also differ across species. We reproduce these biological properties by varying neural resource constraints in a deep convolutional model of the visual system. more on Twitter...
The emergence of multiple retinal cell types through efficient coding of natural movies
Samuel A. Ocko*, Jack Lindsey*, Surya Ganguli, Stéphane Deny
Advances in Neural Information Processing Systems (NeurIPS 2018) / Code / PDF
Why are there so many parallel pathways at the output of the retina? Convolutional autoencoders, that optimally encode natural movies with low firing rates, use exactly the same channels that primates use in their retina. more on Twitter...
Multiplexed computations in retinal ganglion cells of a single type
Stéphane Deny, Ulisse Ferrari, Emilie Macé, Pierre Yger, Romain Caplette, Serge Picaud, Gašper Tkačik & Olivier Marre
Nature Communications 2017 / PDF
By fitting convolutional neural neworks to retinal responses, we find that certain cell types multiplex linear and non-linear computations with the help of a sophisticated gain control mechanism.
Optogenetic vision restoration with high resolution
Ulisse Ferrari*, Stéphane Deny*, Abhishek Sengupta*, Romain Caplette, José-Alain Sahel, Deniz Dalkara, Serge Picaud, Jens Duebel, Olivier Marre
Using Bayesian methods, we estimate an upper bound on the acuity that blind people could recover from optogenetic therapy.
Learning stable representations in a changing world with on-line t-SNE: proof of concept in the songbird
Stéphane Deny*, Emily Mackevicius*, Tatsuo Okubo, Gordon Berman, Joshua Shaevitz, Michale Fee
International Conference on Learning Representations Workshop Track (ICLR 2016) / PDF
By adapting t-SNE to streaming time series, we track the evolution of syllables in the developing songbird.
All Publications
A Unified Theory of Early Visual Representations from Retina to Cortex through Anatomically Constrained Deep CNNs
Lindsey J. *, Ocko S. *, Ganguli S., Deny S. International Conference on Learning Representations (ICLR 2019) / Code / PDF
The emergence of multiple retinal cell types through efficient coding of natural movies
Ocko S. *, Lindsey J. *, Ganguli S., Deny S. Advances in Neural Information Processing Systems (NeurIPS 2018) / Code / PDF
Optogenetic vision restoration with high resolution
Ulisse Ferrari*, Deny S. *, Sengupta A.*, Dalkara D., Duebel J., Marre O. BiorXiv 2018 / PDF
Separating intrinsic interactions from extrinsic correlations in a network of sensory neurons
Ferrari U., Deny S., Chalk M., Tkačik G., Marre O., Mora T. Physical Review E 2018 / PDF
A simple model for low variability in neural spike trains
Ferrari U., Deny S., Marre O., Mora T. Neural Computation 2018 / PDF
Nonlinear decoding of a complex movie from the mammalian retina
Botella-Soler V., Deny S., Marre O., Tkacik G. PLoS Computational Biology 2018 / PDF
A spike sorting toolbox for up to thousands of electrodes validated with ground truth recordings in vitro and in vivo
Yger P., Spampinato G.L.B, Esposito E., Lefebvre B., Deny S., Gardella C., Stimberg M., Jetter F., Zeck G., Picaud S., Duebel J., Marre O. eLife 2018 / PDF
Multiplexed computations in retinal ganglion cells of a single type
Deny S., Ferrari U., Botella-Soller V., Caplette R., Yger P., Tkacik G., Marre O. Nature Communications 2017 / PDF
Learning stable representations in a changing world with on-line t-SNE: proof of concept in the songbird
Deny S.*, Mackevicius E.*, Okubo T., Berman G., Shaevitz J., Fee M. International Conference on Learning Representations Workshop Track (ICLR 2016) / PDF
Dynamical criticality in the collective activity of a population of retinal neurons
Mora T., Deny S., Marre O. Physical Review Letters 2015 / PDF
*equal contribution.
Theses
Local and Non-local Processing in the Retina (PhD thesis)
Deny S. Paris-Sorbonne University VI (2017) / PDF
Surprise Decoding in the Retinal Activity (MS neuroscience thesis)
Deny S. Pierre and Marie Curie University (2013) / PDF
Implementation of a bio-inspired recurrent convolutional neural network for image recognition and threat detection (MS computer science thesis, in french)
Deny S. Ecole Superieure d'Electricite (Supelec) (2012) / PDF