Inria is opening a fixed-term research assistant position on private machine learning for
speech processing as part of the French national collaborative project ANR DEEP-PRIVACY.
The successful candidate will be part of the Magnet team in Lille, which focuses on
privacy-friendly, decentralized and/or graph-based machine learning. He/she will work in
tight collaboration with the Multispeech team in Nancy, which focuses on automatic speech
recognition and speech synthesis.

The goal of DEEP-PRIVACY is to learn automatic speech recognition (ASR) systems from
speech representations that hide speaker identity. Additional requirements such as
decentralized learning, gender fairness, or data efficiency may be considered.

Depending on her/his profile, the candidate will address the following research questions:
- how to design privacy attacks on ASR and design counter-measures;
- how to learn private representations of speech from adversarial training with many
attackers;
- how to design fair and private speech representations;
- how to adapt such methods in the decentralized setting (e.g. federated or fully
decentralized learning);
- how to formally define and measure the trade-off between accuracy, fairness, or privacy
at the global and individual levels.

Application deadline: applications will be assessed on a rolling basis; please apply as
soon as possible.

Starting date: November 1, 2021 or later
Duration: 1.5 year (renewable)
Location: Lille, France
Salary: from 2,050 to 2,130 net/month, according to experience

For more details and to apply:
https://jobs.inria.fr/public/classic/en/offres/2021-04039 (for MSc graduates)
https://jobs.inria.fr/public/classic/en/offres/2021-04034 (for PhD graduates)