Amélie Héliou

Amélie Héliou is a researcher whose career connects bioinformatics, game theory, and AI. She studied bioinformatics at École Polytechnique and earned a doctorate at Université Paris-Saclay before moving through Grenoble's LIG and Criteo AI Lab. Her papers address delayed-reward learning and non-convex optimization. In 2025 she contributed to Magistral, a reasoning model, and in 2026 she coauthored Voxtral TTS, a multilingual speech-generation model.

Now at: Criteo
Works: 39
Citations: 1,660
h-index: 9
Badges: Bridge

Publication numbers from OpenAlex, as of 18 October 2026.

Timeline

By 2018

Bioinformatics training

She studies bioinformatics at École Polytechnique and completes a bioinformatics PhD at Université Paris-Saclay by 2018.
By 2018

A game theory postdoc

She completes a postdoc on game theory at Grenoble's LIG by 2018.
2018

Criteo AI Lab research

In 2018, she joins the Criteo AI Lab research team in Grenoble and works on incrementality.
2020

Learning with delayed rewards

In 2020, she and her coauthors introduce a gradient-free policy for games with delayed rewards and show convergence to Nash equilibrium even when feedback delays are unbounded.
2021

Hierarchical dual averaging

In 2021, she and her coauthors propose hierarchical dual averaging for zeroth-order online non-convex optimization.
June 2025

Magistral reasoning model

In June 2025, she contributes to Magistral, a reasoning model trained with reinforcement learning on Mistral Medium 3.
March–April 2026

Voxtral TTS paper

In March 2026, she coauthors Voxtral TTS, an expressive multilingual text-to-speech model that generates natural speech from as little as three seconds of reference audio.