Alice Coucke

machine learning scientist

I am the head of machine learning Research at Sonos, focusing on privacy-first speech recognition for voice experiences. I was previously the director of machine learning research at Snips (acquired by Sonos in November 2019) that I joined in February 2017.

Prior to joining Snips, I was a PhD student at the theoretical physics lab of the École Normale Supérieure (Paris). I focused on high-dimensional inference with graphical models in the context of protein structure prediction.

Contact: coucke [at] phare [dot] normalesup [dot] org
Twitter: @alicecoucke

ResearchOutreachCV

Outreach

Outreach

Outreach

CV

Research

CURRENT

At Sonos (and previously Snips), we consider the problem of spoken language understanding on small devices typical of IoT applications. We work on designing embedded, private-by-design SLU systems with performance on par with cloud-based commercial solutions. I filed 3 patents currently under review. I have worked on several topics in automatic speech recognition: keyword spotting in the context of wake word detectors, natural language understanding, sentence generation, etc.



  • S. D'Ascoli, A. Coucke, F. Caltagirone, A. Caulier, and M. Lelarge
    Conditioned Query Generation for Task-Oriented Dialogue Systems
    November 2019, arxiv preprint

  • A. Saade, A. Coucke, A. Caulier, J. Dureau, A. Ball, T. Blüche, D. Leroy, C. Doumouro, T. Gisselbrecht, F. Caltagirone, T. Lavril, and M. Primet
    Spoken Language Understanding at the Edge
    October 2019, accepted for oral and poster presentation at the Energy Efficient Machine Learning and Cognitive Computing workshop at NeurIPS 2019, arxiv preprint

    A. Coucke, M. Chlieh, T. Gisselbrecht, D. Leroy, M. Poumeyrol, and T. Lavril
    Efficient Keyword Spotting using Dilated Convolutions and Gating
    November 2018, accepted for publication to ICASSP 2019, paper

  • D. Leroy, A. Coucke, T. Lavril, T. Gisselbrecht, and J. Dureau
    Federated Learning for Keyword Spotting
    October 2018, accepted for publication to ICASSP 2019, paper

  • A. Coucke, A. Saade, A. Ball, T. Blüche, A. Caulier, D. Leroy, C. Doumouro, T. Gisselbrecht, F. Caltagirone, T. Lavril, M. Primet, and J. Dureau
    Snips Voice Platform: an embedded Spoken Language Understanding system for private-by-design voice interfaces
    May 2018, accepted in the spotlight session of the Privacy in Machine Learning and Artificial Intelligence worskhop at ICML 2018, arxiv  preprint


PHD

I pursued my PhD under the supervision of Rémi Monasson and Martin Weigt at the lab of theoretical physics of the École Normale Supérieure (Paris). I focused on maximum entropy graphical model inference and its application to genomic data, especially protein structure prediction. I discussed the extension of these approaches to other challenging fields, such as sequence folding prediction and homology detection. Through an extensive study on both artificial and biological data, I provided a better interpretation of the central inferred parameters, up to now poorly understood. I presented a new and more precise procedure for the inference of generative models, which lead to further improvements on real, finitely sampled data.

  • A. Coucke
    Statistical modeling of protein sequences beyond structural prediction: High-dimensional inference with correlated data
    October 2016, PhD thesis

  • A. Coucke, G. Uguzzoni, F. Oteri, S. Cocco, R. Monasson, and M. Weigt
    Direct coevolutionary couplings reflect biophysical residue interactions in proteins
    November 2016, paper

  • J. Barton, E. De Leonardis, A. Coucke, and S. Cocco
    ACE: adaptive cluster expansion for maximum entropy graphical model inference
    May 2016, paper

  • F. Rizzato, A. Coucke, E. De Leonardis, J. Barton, J. Tubiana, R. Monasson, and S. Cocco
    Inference of compressed Potts graphical models
    July 2019, arxiv preprint


OTHER

In January 2018, I took part in the DAT-ICU datathon for intensive care, aiming at presenting a clinical project using the MIMIC database (data about 50 000 intensive care unit patients). We proposed to associate a clinical print to each patient stay, allowing for instance to identify clusters of similar patients in a non supervised approach. We presented an algorithm of clinical data (textual and numerical) dimensionality reduction based on deep learning techniques.
My team won the first prize.

  • J. Escudié, A. Saade, A. Coucke, and M. Lelarge
    Deep representation for patient visits from electronic health records
    March 2018, arxiv preprint

During my master 2, I worked at the "Physico-Chimie" lab of the Institut Curie under the supervision of Jean-François Joanny. I developed a theoretical model based on active gel theory and performed numerical simulations to study the effect of cell migration on morphogenesis in biological tissues. 

  • E. Hannezo, A. Coucke, and JF. Joanny
    An interplay of migratory and division forces as a generic mechanism for stem cell patterns
    December 2015, paper

I enjoy giving talks and discussing new ideas with people. I am very much interested in finding ways to present and apprehend complicated theories. I could not agree more with the great physicist Richard Feynman who said "If you cannot explain something in simple terms, you don't understand it"

On another topic, I am worried about the gender balance issue in machine learning, and I try to raise awareness on that specific subject.


TECHNICAL TALKS / CONFERENCES

Since February 2017:


OUTREACH

  • I sit at the board of the CNC-RIAM as an Artificial Intelligence specialist. RIAM (Recherche et Innovation en Audiovisuel et Multimédia) provides financial grants to innovative R&D projects in production, processing, distribution, and publication of images and audio.
    It is a partnership between CNC (Centre National du Cinema et de l'image animée) and BPI France

  • I am a mentor for WAI Promo, a fantastic brand new program for high school female students who want to study engineering

  • In September 2019, I was invited as an expert in artificial intelligence to participate to the Committee of Equality and Non Discrimination of the Parliamentary Assembly of the Council of Europe in Paris.
    Council of Europe, September 12, 2019

  • In June 2019, I gave a presentation at the Council of Europe in Strasbourg about artificial intelligence and its gender equality implications. I was invited by Cécile Greboval and Caterina Bolognese from the Gender Mainstreaming Team of the CoE.
    Council of Europe, June 13, 2019

  • In April 2019, I participated in a round table called "Comment rendre l'IA plus inclusive ?" with Aude Bernheim and Peggy Pierrot moderated by Elisa Braun at la Gaité Lyrique in the context of the series of events Computer Grrrls - History, Gender, Technology. 
    Gaité Lyrique, April 21, 2019

  • I have recorded a podcast in French about algorithms (for non-specialists), where I try to define and explain the basic concepts, give many examples, and talk about bias with Rachel Nullans from Brains Agency (November 2018),
    podcast"À la découverte des algorithmes"

  • In March 2018, I participated in a round table about artificial intelligence moderated by Aline Richard Zivohlava.
    Rendez-Vous de l'Inspirations: Les Intelligences Artificielles, Paris @Snips, March 29, 2018
    video"Au commencement est l’humain"

PRESS AND AWARDS

EXPERIENCE


  • Head of Machine Learning Research, Sonos
    November 2019 -- present
    Leading the research & publication effort in Machine Learning.


  • Director of Machine Learning Research, Snips
    August 2019 -- November 2019
    Leading the research & publication effort in Machine Learning.

  • Senior Machine Learning Scientist, Snips
    February 2017 -- August 2019
    machine learning team
    At Snips, we consider the problem of spoken language understanding on small devices typical of IoT applications. We work on designing embedded, private-by-design SLU systems with performance on par with cloud-based commercial solutions. I filed 3 patents currently under review. I worked on the following topics:

    - Designing small, fast, and accurate models to extract the intent and relevant informations from a phonetic representation of a spoken query given by an Acoustic Model and conducting research to improve the end-to-end, speech-to-meaning accuracy of these models.
    - Designing small, fast, and accurate models to detect a predefined keyword in a continuous audio stream and conducting research to improve the accuracy and computational efficiency of these models.
    - Leading the effort in publishing scientific papers (2 at ICASSP 2019, 1 at a workshop at ICML 2018, see Research section) and contributing to the community by open-sourcing code.
    - Benchmarking these models in real-world settings.
    - Implementing and maintaining backend services allowing a fast training of such models though a web-based console.
    - Implementing the core algorithms allowing them to run in real time on devices such as the Raspberry Pi 3, with small memory and computational power.

  • CNC-RIAM board member
    May 2018 -- present
    artificial intelligence referent
    I sit at the board of the CNC-RIAM. RIAM (Réseau de recherche en Innovation et Multimedia) provides financial grants to innovative R&D projects in production, processing, distribution, and publication of images and audio. It is a partnership between CNC (Centre National du Cinéma et de l'image animée) and BPI France}.

  • Data Scientist, Withings
    October 2016 -- February 2017
    platform team
    I worked on improving the wake and sleep detection and the sleep stage detection algorithms of Withings’ connected alarm clock. I explored users clustering with respect to their fixed goals through health behavior change theories.

  • PhD student, Laboratoire de Physique Théorique, École Normale Supérieure
    September 2013 -- September 2016 (defended on October 10, 2016). See the Research section for more details.

  • Research intern, Laboratoire Physico Chimie, Institut Curie, Paris
    January 2013 -- March 2013. See the Research section for more details.

  • Visiting Research Scholar, American Museum of Natural History, New York City
    February 2012 -- August 2012
    I worked on computational astrophysics under the supervision of Mordecai-Mark Mac Low to study supernova-driven turbulence in the interstellar medium.



EDUCATION

  • PhD in statistical physics, École Normale Supérieure
    Defended on October 10, 2016
    under the supervision of Rémi Monasson & Martin Weigt
    See the Research section for more details.

  • Master 2, International Center for Fundamental Physics, École Normale Supérieure
    Mention Bien
    September 2012 -- June 2013
    Theoretical Physics, focus on Mathematics and Statistical Physics.

  • Undergraduate program, École Normale Supérieure
    September 2010 -- June 2013
    promotion Physics 2010
    Undergraduate curriculum with a focus on Mathematics and Theoretical Physics.

  • Classes préparatoires, MPSI-MP*, Lycée Henri IV
    September 2007 - June 2010

Pictures by Alaa Saade
Template created by @lexoyo with Silex, free website builder

Pictures by Alaa Saade
Template created by @lexoyo with Silex, free website builder

Pictures by Alaa Saade
Template created by @lexoyo with Silex, free website builder