Hadrien Hendrikx
 |
BrieflyI am a Research Scientist (Chargé de Recherche) in the Thoth team at Inria Grenoble, a French public research institute. From October 2021 to December 2022, I was a post-doc in the MLO team from EPFL, working with Martin Jaggi. Before that (2018-2021), I was a Ph.D. student in the SIERRA and DYOGENE (now ARGO) teams from Inria Paris, which are also part of the Computer Science Department of Ecole Normale Supérieure. I was also part of the MSR-INRIA joint centre. I worked under the supervision of Francis Bach and Laurent Massoulié on decentralized optimization. Prior to that, I graduated from Ecole Polytechnique in 2016 and got a master degree from EPFL in Computer Science (Master en Informatique) in 2018. During my master, I had the chance to work as a Research Assistant in the DCL lab under the supervision of Rachid Guerraoui and in close collaboration with Aurélien Bellet. Contact
|
Openings
Please reach out by email if you're interested in collaboration!
Research interests
I am broadly interested in optimization for machine learning, regardless of the flavor: stochastic, accelerated, non-euclidean… My PhD mainly focused on decentralized methods for distributed optimization, and in particular how to efficiently leverage acceleration and variance reduction in a decentralized setting. My work leverages principled reformulation-based approaches, that obtain decentralized algorithms (with guarantees) by applying standard (single-machine) optimization theory to well-chosen problems. Efficient algorithms can then be obtained by going back and forth between the reformulations and the optimization tools.
I am more generally open to any problem related to making many entities work together efficiently, potentially without a central authority, and hopefully with some guarantees for the participants. This leads me to read about differential privacy issues in ML, and reinforcement learning theory.
Teaching
2018 - 2019: Teaching assistant, Advanced Algorithms (L3 Informatique), Logic (L1 Informatique) University Paris Descartes
2019 - 2020: Teaching assistant, Advanced Algorithms (L3 Informatique), University Paris Descartes
Reviewing
Conferences: ICML 2019 (Top 5%), NeurIPS 2019 (Top 400), ICML 2020 (Top 33%), NeurIPS 2020 (Top 10%), AISTATS 2021, ICML 2021, NeurIPS 2021, AISTATS 2022, ICML 2022 (Top 10%), NeurIPS 2022, AISTATS 2023 (Top 10%).
Journals: Mathematical Programming, IEEE Transactions on Signal Processing, Automatica, SIOPT, JMLR
Students Supervision
Current students:
Renaud Gaucher: full-time Master intern, co-supervised with Aymeric Dieuleveut
Daniel Morales Broton: full-time Master intern.
Former students:
Abdellah El Mrini: Master Student at Ecole Polytechnique (Paris)
Rustem Islamov: Master Student at Ecole Polytechnique (Paris)
Mathieu Even: PhD student at INRIA Paris with Laurent Massoulié
Publications and preprints
T. Vogels, H. Hendrikx, M. Jaggi. Beyond spectral gap (extended): the role of the topology in decentralized learning.
[arXiv:2301.02151], arXiv preprint, 2023.
H. Hendrikx. A principled framework for the design and analysis of token algorithms.
[arXiv:2205.15015], International Conference on Artificial Intelligence and Statistics (AISTATS), 2023.
T. Vogels, H. Hendrikx, M. Jaggi. Beyond spectral gap: the role of the topology in decentralized learning.
[arXiv:2206.03093], Advances in Neural Information Processing Systems (NeurIPS), 2022.
H. Hendrikx. Accelerated Methods for Distributed Optimization.
[HAL link], PhD Thesis, PSL Research university, 2021.
M. Even, H. Hendrikx, L. Massoulié. Decentralized Optimization with Heterogeneous Delays: a Continuous-Time Approach.
[arXiv:2106.03585], arXiv preprint, 2021.
M. Even, R. Berthier, F. Bach, N. Flammarion, P. Gaillard, H. Hendrikx, L. Massoulié, A. Taylor A Continuized View on Nesterov Acceleration for Stochastic Gradient Descent and Randomized Gossip.
[arXiv:2106.07644], Advances in Neural Information Processing Systems (NeurIPS), 2021. Oral Presentation. Outstanding paper award.
H. Hendrikx, F. Bach, L. Massoulié. An Optimal Algorithm for Decentralized Finite Sum Optimization.
[arXiv:2005.10675], SIAM Journal on Optimization, 2021.
R. Dragomir, M. Even, H. Hendrikx. Fast Stochastic Bregman Gradient Methods: Sharp Analysis and Variance Reduction.
[arXiv:2104.09813], International Conference on Machine Learning (ICML), 2021.
H. Hendrikx, F. Bach, L. Massoulié. Dual-Free Stochastic Decentralized Optimization with Variance Reduction.
[arXiv:2006.14384], Advances in Neural Information Processing Systems (NeurIPS), 2020.
H. Hendrikx, L. Xiao, S. Bubeck, F. Bach, L. Massoulié. Statistically Preconditioned Accelerated Gradient Method for Distributed Optimization.
[arXiv:2002.10726], International Conference on Machine Learning (ICML), 2020.
A. Bellet, R. Guerraoui, H. Hendrikx. Who started this rumor? Quantifying the natural differential privacy guarantees of gossip protocols.
[arXiv:1902.07138], International Symposium on DIStributed Computing (DISC), 2020.
H. Hendrikx, F. Bach, L. Massoulié. An Accelerated Decentralized Stochastic Proximal Algorithm for Finite Sums.
[arXiv:1905.11394], Advances in Neural Information Processing Systems (NeurIPS), 2019.
H. Hendrikx, F. Bach, L. Massoulié. Accelerated Decentralized Optimization with Local Updates for Smooth and Strongly Convex Objectives.
[arXiv:1810.02660], International Conference on Artificial Intelligence and Statistics (AISTATS), 2019.
EME Mhamdi, R Guerraoui, H Hendrikx, A Maurer. Dynamic Safe Interruptibility for Decentralized Multi-Agent Reinforcement Learning.
[arXiv:1704.02882], Advances in Neural Information Processing Systems (NIPS), 2017. Spotlight presentation.
H. Hendrikx, M. Nuñez del Prado Cortez. Towards a route detection method based on detail call records.
[IEEE Xplore 7885725], In Computational Intelligence (LA-CCI), 2016 IEEE Latin American Conference on (pp. 1-6). IEEE., 2016.