Ceci n'est pas Hady Elsahar

Hady
Elsahar

hadyelsahar [at] gmail [dot] com

I'm an AI Research Scientist at Meta AI, specializing in Multilingual, Multimodal Generative Models, I also spend a great deal of time making them accessbile, responsible and safe. My work in AI involves training foundational generative models that scale multilingually and multimodally. I am deeply committed to making AI accessible to everyone, ensuring fairness and representation in technology. Central to my efforts is a focus on removing biases and ensuring safety in AI, balancing cutting-edge innovation with ethical responsibility.

🌟 My research contributions were marked by publications in renouned conferences like Neurips, ICML, and ICLR. My work in Meta AI on foundational models like SeamlessM4T has earned recognition from Time Magazine as one of the Best Inventions of 2023.

Prior to Meta I was a research scientist in Naver. My collaboration with Marc Dymetman was a transformative experience to my growth as a researcher. Our work, on distributional RLHF and Energy Based Models was published in leading conferences like Neurips, TMLR, ICML, and ACL.

Prior to Naver, I was a research intern at Bloomberg London, IBM Germany, and Microsoft, focusing on NLP-related topics.

I am grateful for my educational foundation from L'Université de Lyon, where my Ph.D. journey under the guidance of Christophe Gravier and Frederique Laforest profoundly shaped my research path. My thesis delved into Data2Text Generation and Relation Extraction & Discovery.

Beyond research, I am deeply invested in community engagement and advocacy. I've organized significant workshops like AfricaNLP and contributed to panels on diversity at NAACL. Together with the Masakhane community we have received the Wikimedia Foundation Research Award that highlights our passion for making technology accessible and beneficial for all.

I grew up in the vibrant heart of Cairo, a city in Africa rich with untold stories 🌍. Growing up there, I've experienced firsthand the challenges that come with being part of an underprivileged group in the field of AI research.

🌟 Need Mentorship?: If you are a young researchers out there from under-represented groups in AI, feeling uncertain about your path and want a career advice: don't hesitate to reach out. If you're looking for guidance, a friendly chat, or just a bit of encouragement, drop me a message. I'm more than happy to lend a supportive hand.

Hadyelsahar_Tr4k+%281%29.jpg

لغتي الأم هي العربية. and I have near native speaker knowledge of English, et je dispose de connaissance avancée en français. Ich habe ein grundlegendes Verständnis der deutschen Sprache. 한국어를 모르거나, 이해하는 데 어려움이 있습니다.

🌍 No Borders and Free Identities

In a world where borders are just imaginary lines we imagine 🌐 and identities are ever-changing 🔄 (Borders ∈ 𝕀, Identity ∈ ℂ), I like to think of things a bit differently. Borders? They're like those imaginary lines that everyone makes a big deal out of, much like this last slice of pizza 🍕. The world's already full of these lines; why make more of them? And identity, well, that's a whole different story.

News & Invited Talks

  • 📝 May, 2024: I Audioseal for Audio Watermark against deep fakes is accepted at ICML2024 [paper] [code commercial license] [hugging face] [demo]

  • 🏅 Oct. 24, 2023: SeamlessM4T recognized as one of Time Magazine's Best Inventions of 2023. [Read more]

  • 📝 Aug. 22, 2023: Release of SeamlessM4T, a milestone in Multilingual and Multimodal Machine Translation. [Blog] | [Paper] | [Demo] | [Hugging Face]

  • 🎤 May 5, 2022: Organized the AfricaNLP2023 workshop at ICLR2023, Kigali, Rwanda. [Accepted Papers]

  • 📝 Sep 15, 2022: **Published at Neurips2022 **on Finetuning Language Models. [paper] |

  • 🌟 Sep 5, 2022: Began a new role at Meta AI, focusing on multilingual and multimodal generative models.

  • 📝 May 15, 2022: Paper accepted at ICML2022 on Controlling Conditional Language Models. [Paper]

  • 🎤 April 29, 2022: Presented at the M3L workshop, ICLR2022 on bridging the language gap on Wikipedia.

  • 🎤 April 29, 2022: Organized the 3rd Africanlp workshop at ICLR2022. [Accepted Papers]

  • 📝 Nov 20, 2021: Two papers accepted at CtrlGen workshop, Neurips2021. Topics: EBMs and discrete sampling.papers

  • 🎤 August 6, 2021: Panel participation at the GEM Workshop, ACL2021 and a paper presentation at the NL4Prog workshop. [Paper]

  • 🎤 August 4, 2021: Featured in the Naver Labs Europe Podcast discussing Energy-Based Models. [Recording & Transcript]

  • 🎤 June 7, 2021: Panelist at NAACL2021 on Inclusivity in Conferences.

  • 🎤 June 6, 2021: Spoke at the Black in AI social, NAACL2021, on African language diversity in NLP.

  • 📝 May 3, 2021: Active participation at ICLR2021 - **Oral presentation **on LLM distributional finetuning and AfricaNLP workshop co-organization.

  • 🏅 Apr 14, 2021: Awarded the Wikimedia Foundation Research Award for the Masakhane paper. [Award]

  • 🎤 Apr 2, 2021: Talk at ML Collective DLCT on Controlled Text Generation.

  • 🎤 Mar 3, 2021: Lecture at RECITAL, Paris on Controlling Large Language Models.

  • 🎤 Feb 26, 2021: Course lecture on Neural Language Generation at DSBA master, CentraleSupélec, Paris. [Slides]

  • 📝 Jan 15, 2021: ICLR2021 Oral Presentation: A Distributional Approach to Controlled Text Generation. [Paper]

  • 🎤 Dec 18, 2020: Co-organized the Energy-Based Models Workshop at ICLR 2021. [Workshop Website]

  • 🎤 Dec 15, 2020: Co-organized AfricaNLP Workshop at EACL 2021. [Workshop Website]

  • 📝 Nov 7, 2020: **EMNLP2020 paper **on Participatory Research in Machine Translation. [Paper]

  • 🎤 Apr 3, 2020: **Talks at Bloomberg London and UCL **on Predicting Machine Learning Model Failures. [Slides]

Selected Research Papers

For the full list check my Google Scholar or DBLP

 Blog posts & podcasts