Hédi Hadiji

I am a post-doc in the team of Tim van Erven at the Korteweg-de Vries Institute for Mathematics, of the University of Amsterdam, where I work on adaptive online learning.

I defended my PhD in 2020 at the LMO of the Université Paris-Saclay, studying multi-armed bandits with a focus on adaptation problems, under the supervision of Gilles Stoltz and Pascal Massart. Before that, I graduated from the École Polytechnique in 2016. I completed the Part III of the Mathematical Tripos at the University of Cambridge (2015-2016) and also obtained a Master's degree (M2) in Probability and Statistics at Orsay in 2016-2017.

Contact me at: h dot hadiji at uva dot nl (or at) hedi dot hadiji at gmail dot com


Research

Papers


Scale-free Unconstrained Online Learning for Curved Losses, with Tim van Erven and Jack Mayo, 2022, under review [arxiv]

Between Stochastic and Adversarial Online Convex Optimization: Improved Regret Bounds via Smoothness, with Tim van Erven, Cristóbal Guzmán, and Sarah Sachs, 2022, under review [arxiv]

Distributed Online Learning for Joint Regret with Communication Constraints, with Tim van Erven and Dirk van der Hoeven, 2021, accepted to ALT'22 [arxiv]

Diversity-preserving K-armed bandits, revisited, with Sébastien Gerchinovitz, Jean-Michel Loubes, and Gilles Stoltz, 2020, [HAL]

Adaptation to the Range in K-Armed Bandits, with Gilles Stoltz, 2020, under review [HAL]

Polynomial Cost of Adaptation for X-armed Bandits, Advances in Neural Information Processing Systems (NeurIPS), 2019 [arxiv, poster]

KL-UCB-switch: Optimal Regret Bounds for Stochastic Bandits from both a Distribution-dependent and a Distribution-free Viewpoints, with Aurélien Garivier, Pierre Ménard and Gilles Stoltz, 2018, under review [arxiv]

Talks


Decentralized Online Convex Optimization, March 2022, KdVI Statistics seminar, University of Amsterdam

Decentralized Online Convex Optimization, Feb 2022, Séminaire de Mathématiques appliquées du Laboratoire Jean Leray, Université de Nantes

Fast Convergence to Equilibria in Repeated Games via Online Learning (expository presentation), May 2021, CWI Machine learning weekly seminar

Polynomial Cost of Adaptation for X-armed bandits, February 2020, Thematic Statistics Seminar in Machine Learning, Leiden University

Polynomial cost of adaptation for X-armed bandits, October 2019, Seminar of the Institut für Mathematische Stochastik (IMST), Otto-von-Guericke-Universität, Magdeburg.

Polynomial cost of adaptation for X-armed bandits, September 2019, Workshop of the Celeste INRIA team

X-armed bandits, June 2019, Séminaire des doctorants, Orsay

KL-UCB-switch, May 2018, 50èmes Journées de Statistique, EDF Lab Paris Saclay

Other


My PhD thesis, and the slides from my defense.

Reviewing services: COLT('22, '21), UAI '21, Neurips'19 (top 400 reviewer), Mathematics of Operations Research, SIMODS.

I enjoy implementing the algorithms I design: my personal python bandit laboratory for numerical experiments.


Teaching

I am TAing for the Machine Learning Theory class of the MasterMath program (2021 and 2022).

For the year 2019-2020, I was a teaching assistant for the following classes at Orsay:

  • Mathematical Statistics (M1, 1st year Master's students) [webpage]
  • Statistics for Biologists, including R lab sessions (L2, 2nd year students)

During the years 2017-2019, at the IUT de Sceaux. I TA'ed for the courses:

  • Mathematics for Business, Mathematics and Personal Finance 1st year students
  • Elementary Statistics, 2nd year students

Interests

I enjoy bouldering and I'd like to be better at chess. I read a lot of comics/graphic novels, and sometimes real books. When I feel free, I like to go to the movies.

My first name is Hédi and my last name is Hadiji. In Arabic my first name is written هادي. In English (and in French) pronounce my name like "Eddy".