Draft:Julien Launay

Julien Launay
Alma mater	École normale supérieure Paris-Saclay (MSc) École Polytechnique (MSc) PSL Research University (PhD)
Occupations	Computer scientist, technology entrepreneur
Organization	Adaptive ML
Known for	BLOOM (BigScience) Falcon LLM RefinedWeb dataset
Title	Chief executive officer

Julien Launay is a French computer scientist, researcher, and technology entrepreneur known for his work in large-scale machine learning and high-performance computing. He is the co-founder and chief executive officer of Adaptive ML, an artificial intelligence company based in New York City and Paris. Prior to founding Adaptive ML, Launay held research roles at the Technology Innovation Institute (TII), where he worked on the Falcon series of open-source large language models.

Early life and education

Launay was educated at the École normale supérieure Paris-Saclay (formerly ENS Cachan), where he studied from 2015 to 2019 and obtained a Master of Science in civil engineering. From 2017 to 2018, he served as a visiting research student at the City University of Hong Kong, where he applied machine learning to solar engineering. While completing his studies at ENS Paris-Saclay, Launay also attended École Polytechnique, where he obtained a second Master of Science in climate science in 2019.

He later completed an industrial PhD in applied mathematics at PSL University, titled “Principled modeling methods and beyond backpropagation approaches for the large-scale era.” The research was conducted at LightOn in collaboration with the Laboratoire de Physique de l’École Normale Supérieure (ENS Paris), and explored alternatives to backpropagation such as Direct Feedback Alignment (DFA) using optical co-processors for large-scale matrix multiplication.

Career

Early research and BigScience

Early in his career, Launay worked at LightOn, a French hardware-AI company, and later at Hugging Face. He subsequently participated in the BigScience project and served as chair of the Architecture and Scaling group during development of the multilingual BLOOM model. He also worked on the Falcon family of large language models, including Falcon 40B and Falcon 180B. This work included contributions to the RefinedWeb dataset, a large-scale filtered web corpus used in model training.

Adaptive ML

In 2023, Launay co-founded Adaptive ML with Baptiste Pannier and Daniel Hesslow. The company develops software intended to help organisations customise and improve large language models using reinforcement learning techniques, moving AI systems from proof-of-concept into production use.

In March 2024, Adaptive ML announced a $20 million seed funding round led by Index Ventures with participation from ICONIQ Capital and Databricks Ventures.

Launay is active in the research community, having served as a reviewer for NeurIPS and ICML.

Selected publications

• Scao, Teven Le; Fan, Angela; Akiki, Christopher; Launay, Julien (2022). "BLOOM: A 176B-Parameter Open-Access Multilingual Language Model". arXiv preprint arXiv:2211.05100.

• Penedo, Guilherme; Malartic, Quentin; Hesslow, Daniel (2023). "The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data Only". Proceedings of the Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS) Datasets and Benchmarks Track.

• Almazrouei, Ebtesam; Alobeidli, Hamdan; Alshamsi, Abdulla (2023). "The Falcon Series of Open Language Models". arXiv preprint arXiv:2311.16867.

• Launay, Julien; Poli, Iacopo; Boniface, François; Krzakala, Florent (2020). "Direct Feedback Alignment Scales to Modern Deep Learning Tasks and Architectures". Advances in Neural Information Processing Systems. 33.

• Launay, Julien; Poli, Iacopo; Krzakala, Florent (2019). "Principled Training of Neural Networks with Direct Feedback Alignment". arXiv preprint arXiv:1906.04554.