[Link] to Technical Disclosure Commons
[PDF] of the document
Often, fonts used in online resources such as websites are downloaded from a server to a local device used for browsing the resource. While typical Latin fonts are relatively small, such as on the order of hundreds of kilobytes, some Latin fonts and, more typically, non-Latin, e.g. Chinese, Japanese, Korean, etc., fonts can be much larger, such as on the order of tens of megabytes. Distributing these fonts in their entirety can result in a slow user experience, and so what is needed is a method for faster distribution of these web fonts.
Generally, the present disclosure is directed to serving font files in topical subsets. In particular, in some implementations, the systems and methods of the present disclosure can include or otherwise leverage one or more machine-learned models to predict topic labels of characters in a font based on a corpus of characters or glyphs.
This font is the sliced version of Noto Sans SC , using the topic model + genetic algorithm approach describe here. It serves all 44,683 characters of this font, while Noto Sans SC only serves a subset of 8105 characters .
Did you know that Google Fonts now supports Korean, with 23 choices? Learn the technical details that made this possible on the Google Developers blog 👉 https://t.co/4Xg6nOyOK2 pic.twitter.com/mpC0QMUf1L— Google Fonts (@googlefonts) April 4, 2018