[{"data":1,"prerenderedAt":205},["ShallowReactive",2],{"research":3},{"id":4,"title":5,"body":6,"description":196,"extension":197,"meta":198,"navigation":199,"ogImage":200,"path":201,"robots":200,"schemaOrg":200,"seo":202,"sitemap":203,"stem":12,"__hash__":204},"research\u002Fresearch.md","Research",{"type":7,"value":8,"toc":188},"minimark",[9,13,26,29,34,87,89,93,127,129,133,144,166,168,172,175,185],[10,11,5],"h1",{"id":12},"research",[14,15,16,17,21,22,25],"p",{},"My research sits at the intersection of ",[18,19,20],"strong",{},"applied mathematics"," and ",[18,23,24],{},"deep learning",", with a focus on making AI systems safe and reliable in the presence of adversaries. The guiding question: how can we prove, and not just hope, that learning systems remain robust?",[27,28],"hr",{},[30,31,33],"h2",{"id":32},"research-interests","Research Interests",[35,36,42,54,65,76],"div",{"className":37},[38,39,40,41],"grid","grid-cols-1","md:grid-cols-2","gap-6",[43,44,47],"card",{"icon":45,"title":46},"i-ph-shield-check-duotone","Adversarially Robust Distributed Learning",[14,48,49,50,53],{},"Distributed training scales learning across many workers, but a single malicious worker can corrupt the model through crafted gradients. I study ",[18,51,52],{},"Byzantine-resilient aggregation rules"," (Krum, coordinate-wise median, trimmed mean) and pre-aggregation schemes that guarantee convergence under a bounded fraction of adversaries.",[43,55,58],{"icon":56,"title":57},"i-ph-compass-duotone","AI Safety & Alignment",[14,59,60,61,64],{},"Building on my internship, I aim to contribute to ",[18,62,63],{},"mathematically grounded safety",": formal verification, robustness certificates, and interpretability tools that give us guarantees about model behavior rather than empirical hope.",[43,66,69],{"icon":67,"title":68},"i-ph-flask-duotone","Secure & Reproducible ML Engineering",[14,70,71,72,75],{},"Research tooling matters as much as theory. I build ",[18,73,74],{},"open-source, reproducible frameworks"," that are documented, tested, and packaged, so that robustness results can be re-run, attacked, and extended by anyone.",[43,77,80],{"icon":78,"title":79},"i-ph-function-duotone","Learning Theory",[14,81,82,83,86],{},"From generalization bounds for two-layer ReLU networks to stochastic optimization under constraints, I enjoy the ",[18,84,85],{},"statistical and mathematical foundations"," that make robustness arguments rigorous.",[27,88],{},[30,90,92],{"id":91},"ongoing-work","Ongoing Work",[43,94,97,112],{"icon":95,"title":96},"i-ph-brain-duotone","M2 Research Internship at CMAP, Ecole Polytechnique",[14,98,99,100,103,104,111],{},"From ",[18,101,102],{},"April to October 2026",", I work under ",[105,106,110],"a",{"href":107,"rel":108},"https:\u002F\u002Felmahdielmhamdi.com\u002F",[109],"nofollow","El Mahdi El Mhamdi"," on robust distributed learning with adversaries: gradient manipulation attacks, open-source research tooling, and improvements to existing aggregation frameworks.",[14,113,114,115,121,122,126],{},"The experimental backbone is ",[18,116,117],{},[105,118,120],{"href":119},"\u002Fprojects\u002Fkrum","Krum",", an open-source framework for Byzantine-resilient aggregation, installable via ",[123,124,125],"code",{},"pip install krum",".",[27,128],{},[30,130,132],{"id":131},"phd-directions","PhD Directions",[14,134,135,136,139,140,143],{},"My PhD at ",[18,137,138],{},"CMAP, Ecole Polytechnique"," starts in ",[18,141,142],{},"November 2026",", continuing this line of work. I plan to explore:",[145,146,147,154,160],"ul",{},[148,149,150,153],"li",{},[18,151,152],{},"Byzantine robustness at scale",": aggregation rules that stay provably safe as clusters grow and decentralization increases.",[148,155,156,159],{},[18,157,158],{},"Formal verification for ML pipelines",": bridging worst-case guarantees and practical training.",[148,161,162,165],{},[18,163,164],{},"Safety-critical applications",": from federated learning to autonomous decision systems.",[27,167],{},[30,169,171],{"id":170},"talks","Talks",[14,173,174],{},"I present my work whenever I get the chance, and slides are linked when available.",[43,176,179],{"icon":177,"title":178},"i-ph-presentation-duotone","Byzantine Robustness in Distributed Learning",[14,180,181,184],{},[18,182,183],{},"September 2026",": Internal Seminar, CMAP, Ecole Polytechnique. Introducing the Krum framework and our adaptive threshold mechanisms for Byzantine-resilient aggregation.",[14,186,187],{},"More talks and publications will appear here as my research progresses.",{"title":189,"searchDepth":190,"depth":190,"links":191},"",2,[192,193,194,195],{"id":32,"depth":190,"text":33},{"id":91,"depth":190,"text":92},{"id":131,"depth":190,"text":132},{"id":170,"depth":190,"text":171},"My research interests in AI Safety, adversarial robustness, and distributed learning, along with my ongoing work at CMAP, Ecole Polytechnique.","md",{},true,null,"\u002Fresearch",{"title":5,"description":196},{"loc":201},"xY9S2VMsCE76AYMYfOBOwCMrQ9ItaCvsz0MI1VI8tZM",1785520008635]