{
  "id": "1707.03684",
  "version": "v1",
  "published": "2017-07-01T05:38:55.000Z",
  "updated": "2017-07-01T05:38:55.000Z",
  "title": "Structured Sparse Ternary Weight Coding of Deep Neural Networks for Efficient Hardware Implementations",
  "authors": [
    "Yoonho Boo",
    "Wonyong Sung"
  ],
  "comment": "This paper is accepted in SIPS 2017",
  "categories": [
    "cs.CV"
  ],
  "abstract": "Deep neural networks (DNNs) usually demand a large amount of operations for real-time inference. Especially, fully-connected layers contain a large number of weights, thus they usually need many off-chip memory accesses for inference. We propose a weight compression method for deep neural networks, which allows values of +1 or -1 only at predetermined positions of the weights so that decoding using a table can be conducted easily. For example, the structured sparse (8,2) coding allows at most two non-zero values among eight weights. This method not only enables multiplication-free DNN implementations but also compresses the weight storage by up to x32 compared to floating-point networks. Weight distribution normalization and gradual pruning techniques are applied to mitigate the performance degradation. The experiments are conducted with fully-connected deep neural networks and convolutional neural networks.",
  "revisions": [
    {
      "version": "v1",
      "updated": "2017-07-01T05:38:55.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "deep neural networks",
      "structured sparse ternary weight coding",
      "efficient hardware implementations"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}