Facebook AI released Opacus, a new PyTorch library for machine learning with differential privacy.

Judge as of 2020-08-31. True: a dated official announcement by Facebook AI (blog post, GitHub release, or PyPI release record) shows that Opacus, a PyTorch library for training machine learning models with differential privacy, was publicly released by Facebook AI on or around that date, and that it was new at that point (not a long-established library). Mostly true: Opacus was released by Facebook AI as a differential-privacy library for PyTorch, but a minor detail is off, such as the release coming somewhat earlier in 2020 or the library being a renamed version of an earlier Facebook project. Mostly false: a library by that name existed but was not a Facebook AI release, or was not a differential-privacy library for PyTorch. False: no such library had been released by Facebook AI as of that date, shown for example by a release record dated later or by attribution to another organisation.

Set before the research began.

1. E1 Introducing Opacus: A high-speed library for training PyTorch models with differential privacy Facebook AI (ai.meta.com blog) · 31 August 2020

   “We are releasing Opacus, a new high-speed library for training PyTorch models with differential privacy (DP) that’s more scalable than existing state-of-the-art methods.”

2. E2 Blog — page 21 — OpenMined OpenMined · 31 August 2020

   “Aug 31, 2020 News Meta Introduces Opacus: A High-Speed Privacy Library for PyTorch with OpenMined Integration”

3. E3 Facebook Open-Sources Machine-Learning Privacy Library Opacus InfoQ · 13 October 2020

   “Facebook AI Research (FAIR) has announced the release of Opacus, a high-speed library for applying differential privacy techniques when training deep-learning models using the PyTorch framework.”

E1 is Facebook AI's blog post dated 2020-08-31. It says Facebook AI is "releasing Opacus, a new high-speed library for training PyTorch models with differential privacy." An OpenMined listing from the same day (E2) reports Meta introducing Opacus as a privacy library for PyTorch, and a later InfoQ report (E3) says Facebook AI Research released Opacus as a differential-privacy library for PyTorch. Together these show a new library released on or around the claim date, which meets the criteria for true.