arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2012.03458cs.LG
收藏

无乘法的深度神经网络训练

Deep Neural Network Training without Multiplications

Preferred Networks, Inc.(Preferred Networks公司)

Tsuguo Mogami

摘要:

深度神经网络真的需要乘法吗?在这里,我们建议只添加两个IEEE 754浮点数与整数相加指令代替浮点乘法指令。我们表明,ResNet可以使用这种操作进行训练,具有竞争力的分类精度。我们的建议不需要任何方法来解决不稳定性和准确性下降,这在低精度训练中很常见。在某些情况下,我们可能会获得与基线FP32结果相同的精度。这种方法将能够消除深度神经网络训练和推理中的乘法运算。

英文摘要:

Is multiplication really necessary for deep neural networks? Here we propose just adding two IEEE754 floating-point numbers with an integer-add instruction in place of a floating-point multiplication instruction. We show that ResNet can be trained using this operation with competitive classification accuracy. Our proposal did not require any methods to solve instability and decrease in accuracy, which is common in low-precision training. In some settings, we may obtain equal accuracy to the baseline FP32 result. This method will enable eliminating the multiplications in deep neural-network training and inference.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑