Yann Dauphin

Dauphin introduced techniques that became standard building blocks in deep learning. Gated Linear Units improved how information flows through sequence models, Mixup made models more robust by training on blended examples, and top-k sampling helped control the quality of generated text.