ProjectionNet: Learning Efficient On-Device Deep Networks Using Neural Projections · HackerPapers discussionshttps://hackerpaper.kaipfister.de/p/1708.00630/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMTProjectionNet: Learning Efficient On-Device Deep Networks Using Neural Projections — mentioned by spiznnxhttps://news.ycombinator.com/item?id=20871644urn:hackerpapers:mention:1708.00630:20871644Tue, 03 Sep 2019 21:58:34 GMT<p>It&#x27;s completely on-device [0]. The prediction models have been shrunk [1] so no connections are made (Google has enabled lots of shrunken on-device models with TF Lite [2]) . I&#x27;m not sure why this page doesn&#x27;t call that out considering how privacy-focused the other features are. 0. https://www.androidauthority.com/smart-reply-android-q-98364... 1. https://www.tensorflow.org/lite/models/smart_reply/overview 2. https://arxiv.org/pdf/1708.00630.pdf, https://www.tensorflow.org/lite/guide</p><p><a href="https://hackerpaper.kaipfister.de/p/1708.00630/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.LG