Welcome to the 15th edition of MLguru, a bi-weekly Machine Learning newsletter. In this edition you will read about:
The State of Machine Learning Frameworks in 2019,
Machine Translation by Facebook,
And much more!
We also still have few places for our face-to-face Machine Learning workshops that will take place in Dublin. Click here for more details.
Are you a Machine Learning pro already? Great! We are hiring for the position of Senior Machine Learning Engineer. Join the team and help us empower international clients like Volkswagen, IKEA or Keller Williams, as well as startups and industry innovators. Visit our job posting to find out how we can help your career.
Machine translation (MT) is one of the most successful applications of natural language processing (NLP) today, outperforming human-level performance in some language translation tasks. Unfortunately, the current high-performing MT systems need large-scale parallel data sets to fit hundreds of millions of model parameters. Most languages today don’t have resources to support this process, and there is a need need to build systems that can work effectively for everyone.
To solve this issue, Facebook AI team have developed a novel approach that combines several methods, including iterative back-translation and self-training with noisy channel decoding, which can tackle this problem. Read more
The new version of BERT is here! exBERT is an interactive tool that provides insights into the meaning of the contextual representations by matchinghuman-specified input to similar contexts in a large annotated dataset. By aggregating the annotations of the matching similar contexts, exBERT helps to give an intuitive explanation of what each attention-head has learned. Read more
PyTorch continues to work on meeting the needs of its users. It has received enthusiastic support from the AI community, and its citations in papers on ArXiv grew by 194 percent in the first half of 2019 alone. Facebook, Microsoft, and Uber are increasingly using it as the foundation for their most important machine learning research and production workloads.
The release of PyTorch 1.3 includes experimental support for features such as seamless model deployment to mobile devices, model quantization for better performance at inference time, and few front-end improvements. Learn more about all the new functionalities on the official PyTorch website.
Hydra is a lightweight framework that simplifies the development of Python applications, especially complex ones, by introducing the ability to compose and override configurations. Hydra’s flexible approach to developing, creating, and maintaining code and configurations can help speed the development of complex applications in various fields, including machine learning research. Read Facebook’s article about all it’s features.