- CES 2021 will take place online. The Consumer Technology Association (CTA) has made it official that their event will not take place in Las Vegas next year as usual. Gary Shapiro, CEO of the Association, told our colleagues from VentureBeat that the fair will however run well from January 6 to 9, 2021 and will allow exhibitors, visitors and the press to participate in keynotes and interviews but online. And Mr. Shapiro even indicates that version 2022 will take place in hybrid version: part in physics, and content online exclusively.

- Microsoft closer to PyTorch. The PyTorch machine learning framework created by Facebook is receiving heavy support. After announcing the DeepSpeed ​​open source project last winter to speed up pyTorch without major code rewriting, Microsoft is investing a little more in this competitive open source framework.t of Tensor Flow from Google. In parallel with the launch of version 1.6, the Redmond firm has announced that it will itself take charge of the development and maintenance of PyTorch builds for Windows. “Lately, a few features for PyTorch were simply not available for the Windows platform, such as distributed training support and the TorchAudio domain library. To alleviate this pain, Microsoft is happy to bring its Windows expertise to the table and bring PyTorch on Windows to its best level, can we read in a PyTorch blog common between Microsoft and Facebook.

- Linkedin switches DeText to open source. Linkedin's Detext framework specializing in natural language-based processing processes to classify, classify and generate tasks is ported to open source. "The framework enables users to better utilize the models and incorporate them into real applications, explained Weiwei Guoo, senior enineering manager at Linkedin. "It's been applied on LinkedIn for search and recommendation ranking, query intent classification, and query auto-completion, with significant improvements in relevance ranking for members searching for people and jobs." . DeText contains several components that can be customized via preloaded templates such as text encoding models and an MLP (multilayer perceptron) layer that can be combined with wide and deep learning functions.