Professor & SCALE-AI Chair at MAGI @polymtl | @ivadolabs advisor Tweets on #ORMS & #Trustworthy #MachineLearning Open-source codes: https://nitter.gq/t.co/HaZVxBWKxX
Montréal, Canada
Joined May 2017
- Tweets734
- Following454
- Followers1.8K
- Likes1.1K
Pinned Tweet
Fresh off the press! Our new survey explores how Combinatorial Optimization (CO) & formal methods help make #MachineLearning more transparent, explainable, fair, robust, and private! arxiv.org/abs/2607.07762
How can we plan orbital debris removal at scale, with multiple spacecraft operating simultaneously? 🚀
We present our work today at #IFORS2026:
🕑 14:15–15:45
📍 JUR – Hörsaal 15
📋 ME-33 — Space Logistics
The debris removal challenge is to jointly decide:
- which debris each spacecraft should target
- in which order & at what time
- using which orbital transfers
These decisions are coupled, as debris motion and orbital-plane drift make transfer costs strongly time-dependent.
Our approach combines an enhanced RQ-law for transfer planning with an adapted Hybrid Genetic Search (HGS) to coordinate the fleet.
Joint work with @TheoGuyard, Mingde Yin, Graham Mackintosh, and Thomas Templin.
#ORMS #SpaceLogistics #Optimization
We hope the paper can serve as a useful guide, and broaden the view of what #ORMS can achieve for #Trustworthy #MachineLearning, as its capabilities are still too often underestimated & dismissed as intractable.
Co-authored with @FerryJulien12, it concludes nearly three years of manual curation, reading, and synthesis. We are submitting now and warmly welcome feedback! I will also be attending #IFORS2026 and am very pleased to discuss this topic!
How much does routing optimization matter?
Compared with Google OR-Tools and Hexaly, HGS-Advance shows substantial gains. In large supply chains, even a 4% reduction in travel distance translates into millions in savings.
ALT Average optimality gap (%) by instance size on the Homberger-Gehring VRPTW benchmark, comparing HGS-Advance with Google OR-Tools 9.14 and Hexaly 14.5. Lower is better.
The benchmark is the Homberger-Gehring VRPTW dataset, with OR-Tools and Hexaly results taken from Hexaly's public comparison: hexaly.com/benchmarks/hexaly…
The votes have been cast: HGS "Advance" is now part of the TIG (@tigfoundation) ecosystem and adopted by benchmarkers! It is now, by a very significant margin, the SOTA for #VRPTW, substantially improving upon DIMACS challenge-winning results and existing commercial solvers.
#TIG’s game-changing innovation model has also been a key motivator. The code takes inspiration from successful ideas in previous contributions, notably `fast_lane_v4` 🙏, and will likewise be built upon by new contributors, sustaining a continuous stream of improvements🚣
Full Professor at Polytechnique Montréal🎉!
As I pursue my research career at this vibrant institution, I am incredibly grateful for this environment and for the community we have built together. Beyond the pursuit of new discoveries, papers, or grants, there is something very special in the daily life of our research group: we share lunch almost every day, "on the 5th floor" within the SCALE-AI Chair team. This is our shared space where we discuss ideas, jokes, challenges, ambitions, doubts, and new directions, and we grow together as a group.
None of the successes we have achieved would have been possible without these bonds. I feel deeply grateful to all the students, colleagues, collaborators, and friends who have been part of this journey.
Looking forward to the next steps of this adventure!
Happy to announce that our paper was rejected as a spotlight (5/5/4) at #ICML2026.
If the methodology was complex enough to confuse the metareviewer, perhaps it may still be of broader interest to you 🙂.
Happy to discuss the work if you are into optimal counterfactual maps that permit explanations in milliseconds, or into the occasional ups and downs of academic publishing 🚣
Paper link for the curious ones: arxiv.org/abs/2602.09128
AI algorithm discovery on #TIG #VehicleRouting challenge at AGI House, SF
27 Claude agents by @StevenDiam77921 and @tigfoundation
Live ≈4h, revisited and rediscovered 30 years of VRP literature.
demo.discoveryatscale.com/
I'm amazed.
