{"author_name":"mnozawa","author_url":"https://blog.hatena.ne.jp/mnozawa/","width":"100%","blog_title":"Ridge-institute R&D Blog","published":"2021-06-30 11:00:00","version":"1.0","image_url":"https://img.esa.io/uploads/production/attachments/9763/2021/06/21/97131/26b52a53-069d-4e73-843b-27846688d90a.png","html":"<iframe src=\"https://hatenablog-parts.com/embed?url=https%3A%2F%2Fiblog.ridge-i.com%2Fentry%2F2021%2F06%2F30%2F110000\" title=\"\u5206\u6563\u5b66\u7fd2\u306b\u304a\u3051\u308b\u540c\u671f\u901a\u4fe1\u30b3\u30b9\u30c8\u524a\u6e1b\u306e\u305f\u3081\u306e\u52fe\u914d\u5727\u7e2e - Ridge-institute R&amp;D Blog\" class=\"embed-card embed-blogcard\" scrolling=\"no\" frameborder=\"0\" style=\"display: block; width: 100%; height: 190px; max-width: 500px; margin: 10px 0px;\"></iframe>","blog_url":"https://iblog.ridge-i.com/","url":"https://iblog.ridge-i.com/entry/2021/06/30/110000","provider_name":"Hatena Blog","provider_url":"https://hatena.blog","description":"\u3053\u3093\u306b\u3061\u306f\uff0e\u682a\u5f0f\u4f1a\u793eRidge-i\u3067\u30a4\u30f3\u30bf\u30fc\u30f3\u3092\u52d9\u3081\u3066\u3044\u307e\u3059mitsuhiko-nozawa\u3067\u3059\uff0e\u672c\u8a18\u4e8b\u3067\u306f\uff0cGradient Compression\u3068\u547c\u3070\u308c\u308b\uff0c\u5206\u6563\u5b66\u7fd2\u74b0\u5883\u4e0b\u3067\u52fe\u914d\u306e\u901a\u4fe1\u91cf\u3092\u524a\u6e1b\u3059\u308b\u624b\u6cd5\u306b\u3064\u3044\u3066\u7d39\u4ecb\u3057\u307e\u3059\uff0e \u306a\u304a\u672c\u8a18\u4e8b\u306f@machinery81\u3068@zawatsky_r\u306b\u3088\u3063\u3066\u30ec\u30d3\u30e5\u30fc\u3055\u308c\u3066\u3044\u307e\u3059\uff0e TL;DR Gradient Compression\u3068\u306f Gradient Compression\u306e\u6982\u8981 Gradient Compression\u306e\u5206\u985e Gradient Sparcification Gradient Quantization Gradient Sparcific\u2026","title":"\u5206\u6563\u5b66\u7fd2\u306b\u304a\u3051\u308b\u540c\u671f\u901a\u4fe1\u30b3\u30b9\u30c8\u524a\u6e1b\u306e\u305f\u3081\u306e\u52fe\u914d\u5727\u7e2e","categories":["gradient compression","distributed deep learning"],"height":"190","type":"rich"}