David Grangier

David Grangier is a machine-learning researcher whose work spans information retrieval, neural translation, question answering and language-model training. He completed an EPFL doctorate in 2008 and held research roles at NEC Labs America, AT&T Research, Microsoft Research, Facebook AI Research and Google Research’s Brain team. His work includes a fully convolutional sequence-to-sequence model and the ELI5 corpus for long-form question answering; his paper on learning convolutional strides received an ICLR Outstanding Paper Award in 2022. As of October 2026, he works with Apple Research’s Machine Learning Group and has coauthored research on specialist language models and memory-augmented pretraining.

Now at: Apple
Works: 131
Citations: 16,717
h-index: 37
Badges: Rising
Bridge
Wayfarer

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

2008

An EPFL doctorate

In 2008, he completes an EPFL PhD with a dissertation titled Machine Learning for Information Retrieval, co-directed by Samy Bengio.
2008–2014

NEC, AT&T and Microsoft roles

He works at NEC Labs America from 2008 to 2011, AT&T Research from 2011 to 2012, and Microsoft Research from 2012 to 2014.
2017

Convolutional sequence modeling

In 2017, he is a research scientist at Facebook AI Research and coauthors Convolutional Sequence to Sequence Learning, an all-convolutional sequence-to-sequence architecture.
2019

ELI5 long-form question answering

In 2019, he works with Google Research’s Brain team and coauthors ELI5: Long Form Question Answering, a corpus of 270,000 Reddit threads for long-form answers.
2022

An ICLR Outstanding Paper

In 2022, Learning Strides in Convolutional Neural Networks receives an ICLR Outstanding Paper Award; the paper introduces DiffStride, a downsampling layer with learnable strides.
2025

Task-adaptive language models

In 2025, he coauthors Task-Adaptive Pretrained Language Models via Clustered-Importance Sampling, an ICLR paper that builds specialist models from generalist training sets by sampling clusters to match the specialist data.
As of October 2026

Language-model research at Apple

He works with Apple Research’s Machine Learning Group in Cupertino. In 2026, he coauthors Pretraining with Hierarchical Memories: Separating Long-Tail and Common Knowledge, which proposes a memory-augmented architecture and pretraining strategy for language models.