Resume

Lisheng Sun-Hosoya
LISN, Université Paris Saclay
Rue Raimond Castaning Bât 660
91190 Gif-sur-Yvette France
Email: sun-hosoya@chalearn.org OR cecile829@gmail.com

I am a Postdoctoral Researcher at LISN, Université Paris-Saclay, since December 2022, with responsibilities in both research and teaching (chargée de recherche et d’enseignement). My current research focuses on AutoML, Meta-Learning, and AI fairness.

Prior to my postdoc, I was a Senior Data Scientist and Research Lead at Crédit Agricole S.A., where I specialized in AutoML, Explainable AI (XAI), and AI fairness for the banking sector.

I earned my Ph.D. in Machine Learning in December 2019 under the supervision of Prof. Isabelle Guyon and Prof. Michèle Sebag, focusing on AutoML and Meta-Learning.

Professional experiences

• Dec. 2022 - Present: Postdoc Researcher, Chargée de Recherche et d’enseignement, LISN, France. Conducted research on machine learning and data bias, resulting in multiple publications. Supervised master’s students on their M1 and M2 projects, leading to a conference publication, and guided research projects with PhD students, resulting in a journal publication and a submission to DMLR. Additionally, taught courses on AI challenges and optimization, with a textbook currently in preparation.

• Jan. 2020 - Nov. 2022: Senior Data scientist, Rérérent de Recherche, Crédit Agricole S.A., France. Designed and implemented innovative AI pipelines for banking systems, including the generation of financial indicators and the architecture of solutions utilizing AutoML and Meta-Learning to enhance learning efficiency. Made significant contributions to R&D project management, particularly in areas such as Green AI, Explainable AI, and Reinforcement Learning. Additionally, supervised internships and facilitated scientific communication with regional branches.

• 2017: Software engineer. Collaboration with Google Zurich (, project manager) on prepararing the AutoDL challenge, Zurich, Switzerland and ChaLearn, California, USA. Contributions to the design and implementation of an AutoDL starting kit using [Tensorflow] [Github].

• 2017: Software engineer. See4C EU project, Forecasting the French electricity power grid (, supervisor). Université Aix-Marseille, France. Contributions to the design and implementation of a starting kit in Python to treat problems of spatio-temporal time series [Github].

• 2016: Software engineer. Codalab Worksheets. Prototype use cases of worksheets in competitions organized on the Codalab platform (, project manager). Implement five worksheets to boost collaboration and use of re-usable workflows, including the BeatAutoSklearn hackathon, run at NeurIPS 2016 challenges in machine learning workshop.[dockerhub]

• 2014-2015: Freelance Software engineer. Japan.

• 2013: Research projet engineer in Astrophysics. Observatoire de Paris, France.

Education

Ph.D. Machine Learning. Université Paris Sud, Gif-sur-Yvette, France, 2019.

Meta-Learning as a Markov Decision Process.
Supervisors: Prof. Isabelle Guyon and Michèle Sebag

The success of machine learning applications across various fields depends heavily on the design of effective model architectures, a process that is both time-consuming and requires substantial domain expertise. My PhD thesis focuses on developing a methodology to automate this process, known as AutoML. In particular, we explore the meta-learning aspect of AutoML and formulate the problem as a Markov Decision Process (MDP). We propose several solutions to this problem, including approaches based on collaborative filtering and reinforcement learning.

Master. Astrophysics. Observatoire de Paris, Paris, France, 2011-2013

The physical properties of local galaxies with high specific star formation rate

Directed by Prof. Matthew Lehnert and Prof. Wim van Driel, Institut d’Astrophysique de Paris, Paris, France.

Bachelor. Physics. Université Paris 7, Paris, France, 2008-2010

Selected Publications

Full Publications List

• Stylized Meta-Album: Multi-domain computer vision meta-dataset, Under review at DMLR, 2024
• RelevAI-Reviewer: A Benchmark on AI Reviewers for Survey Paper Relevance, CAp 2024.
• Meta-learning from Learning Curves Challenge: Lessons from the First Round and Design of the Second Round, ICML 2022 Workshop.
• Analysis of the AutoML challenge series 2015–2018, in AutoML: methods, systems, challenges, Springer, 2019.
• ActivMetal: Algorithm recommendation with Active Meta Learning, ECML PKDD 2018 Workshop.

Student Supervision

• Paulo Couto De Resende Silva, Université Paris Saclay. M1 thesis “LLM based AI-reviewer”, Université Paris Saclay, 2024

• Chiheb Eddine NAJJAR, École des Ponts ParisTech, master’s thesis “Reinforcement Learning For Lead Qualification”, Crédit Agricole S.A., 2021.

• Manh Hung Nguyen, CentraleSupélec, master’s thesis “RL-based Meta-learning from Learning Curves”, Université Paris Sud, 2021. Co-supervised with Prof. Isabelle Guyon.

• Achraf Azize, École Polytechnique, master’s thesis “Interpretability and AutoML”, Crédit Agricole S.A., 2020.

• Nathan Grinsztajn, École Polytechnique, master’s thesis “AutoML and REVEAL games”, Université Paris Sud, 2019. Co-supervised with Prof. Isabelle Guyon.

Teaching experiences

• Advanced Optimization and Automated Machine Learning, M2 Université Paris-Saclay (2022–Present). Co-taught by Lisheng Sun-Hosoya and Solal Nathan, with contributions from Prof. Isabelle Guyon and Marc Schoenauer.

• Creation of an AI Challenge, M1 Université Paris-Saclay (2022–Present). Taught by Lisheng Sun-Hosoya, with contributions from Prof. Isabelle Guyon and Ihsan Ullah. [Link to class under repair following a cyberattack]

• Teaching Assistant Deep Learning in Practice. Mini-projets on Machine learning taught by Prof. Guillaume Charpia, Université Paris Sud, Gif-sur-Yvette, France. I taught the Chapter “AutoML and AutoDL”.

Invited Talks

• International Workshop on Machine Learning and Artificial Intelligence, Machine Learning, Human Learning and Robotics Program, Télécom Paris. Meta-Learning as a Markov Decision Process. Video. (Invited)

• Google Zurich. AutoML1 and its post challenge analysis. (Invited)

• AutoML 2015 workshop @ ICML 2015. Solution to the AutoML challenge. (Invited as the challenge’s 3rd place winner)

Reviewing

• JMLR 2023

• NIPS2018, 2019, 2020

• AutoML@ICML2020, 2021

• MetaLearn@AAAI2021

• IJCNN 2020

Honours and Awards

• $3^{rd}$ place in AutoML 2015-2016 challenge.

Open source project

I put some of my work online for public use.

• Github: https://github.com/LishengSun Research and personal projects.

• Docker: https://hub.docker.com/ I developed several docker images for various projects, for example, this one allows running AutoML challenge participant codes in a reproducible and controllable way.

References

• Prof. Isabelle Guyon: Research Director, DeepMind.

• Anne-Catherine Letournel: Research Engineer, LISN, Université Paris-Saclay.

• Evelyne Viegas: Senior Director, Global Research Engagement at Microsoft Research.

• André Elisseeff: Software Engineering Manager at Google.

• Cécile Capponi: Prof. Univ. Aix-Marseille.

• Michèle Sebag: Directrice de recherche CNRS.