INQ(incremental network quantization：towards lossless CNNs with low-precision weights

incremental network quantization：towards lossless CNNs with low-precision weights 實驗前提：a.一個預訓練的模型，模型已收斂                 b.訓練這個模型訓練集，驗證集，測試集 大體三個步驟：weights partition、group-wise quantization 、re-trainin

>>阅读原文<<