August 2019 Summaries

In Arua, Uganda, university student Samuel Mugisha, motivated by the poor state of vaccination records in his community, took the initiative to create a technological solution with the help of his peers from Developer Student Clubs (DSC). The team developed a mobile app aimed at replacing the unreliable paper records, allowing healthcare workers to track vaccinations accurately and receive timely notifications for upcoming vaccinations. Their innovative approach won the DSC Solution Challenge, a competition that encourages students to tackle local issues using technology. Winning the challenge provided the team with resources and expertise from Google Developer Group Leads, helping them refine their app to function offline, crucial in areas with limited internet access. Despite their success, the team continues to enhance the app by improving its usability and developing a version for parents to monitor their child's vaccinations. Their journey highlights how young developers, supported by platforms like DSC, can address pressing community problems and drive meaningful change.

Prerendering in AMP (Accelerated Mobile Pages) significantly enhances page load times by allowing content to be displayed almost instantly once a link is clicked, thanks to Google's privacy-preserving prerendering techniques. The AMP framework's ability to understand page layout and resource loading status enables it to calculate when "above the fold" content is ready, introducing the custom metric First Viewport Ready (FVR) to measure this speed. Unlike the general First Contentful Paint (FCP) metric, which measures when the first DOM content is rendered, FVR offers a more accurate reflection of when essential content becomes visible by factoring in prerendering. Prerender-adjusted First Contentful Paint (PFCP) further refines this measurement for AMP documents by accounting for prerendering time. While prerendering consistently improves load times, AMP currently remains unique in offering this benefit in a privacy-preserving manner, though future web technologies like Signed Exchanges may extend similar advantages to non-AMP documents.

Coral's recent updates highlight advancements in local AI capabilities, including enhanced post-training quantization support, new APIs for transfer learning, and innovative models optimized for the Edge TPU. The compiler now supports full integer post-training quantization, allowing models to be fully quantized to 8-bit integers, which not only reduces model size but also enables acceleration via Coral's Edge TPU. The Edge TPU Python library update introduces a backpropagation API for near-real-time transfer learning, while the existing imprinting API allows for quick retraining of model classes. Additionally, a new TensorFlow Lite delegate for the Edge TPU lets users accelerate models using the TensorFlow Lite interpreter API. Notably, Coral collaborated with the Edge TPU and AutoML teams to release the EfficientNet-EdgeTPU models, which offer high classification accuracy in a compact form optimized for the Edge TPU. Arrow Electronics provides a student-teacher discount to encourage educational use of Coral boards, and Coral invites ongoing feedback to refine its platform further.