{
  "id": "2106.10796",
  "version": "v1",
  "published": "2021-06-21T01:15:12.000Z",
  "updated": "2021-06-21T01:15:12.000Z",
  "title": "CD-SGD: Distributed Stochastic Gradient Descent with Compression and Delay Compensation",
  "authors": [
    "Enda Yu",
    "Dezun Dong",
    "Yemao Xu",
    "Shuo Ouyang",
    "Xiangke Liao"
  ],
  "comment": "12 pages",
  "categories": [
    "cs.LG",
    "cs.DC"
  ],
  "abstract": "Communication overhead is the key challenge for distributed training. Gradient compression is a widely used approach to reduce communication traffic. When combining with parallel communication mechanism method like pipeline, gradient compression technique can greatly alleviate the impact of communication overhead. However, there exists two problems of gradient compression technique to be solved. Firstly, gradient compression brings in extra computation cost, which will delay the next training iteration. Secondly, gradient compression usually leads to the decrease of convergence accuracy.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2021-06-21T01:15:12.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "distributed stochastic gradient descent",
      "delay compensation",
      "gradient compression technique",
      "parallel communication mechanism method",
      "communication overhead"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 12,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}