DjVu and its connection to Deep Learning (2023)
Summary
Scott Locklin discusses DjVu as a highly efficient document format with strong compression for scanned texts, contrasts it with PDF, and links its origins to the early work of LeCun, Bottou, and Bengio in deep learning. The post covers technical components like IW44 backgrounds, JB2 for text, and ZP-coder arithmetic coding, notes the lack of broad reader support, and reflects on its potential historical impact and personal use on e-readers.