arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1910.04576cs.CV
收藏

高效深度神经网络的训练秩剪枝

Trained Rank Pruning for Efficient Deep Neural Networks

Yuhui Xu, Yuxi Li, Shuai Zhang, Wei Wen, Botao Wang, Wenrui Dai, Yingyong Qi, Yiran Chen, Weiyao Lin, Hongkai Xiong

摘要:

为了加速DNN的推理,低秩近似由于其坚实的理论基础和有效的实现而被广泛采用。之前的几项工作试图通过低秩分解直接近似预训练模型;然而,参数中的小近似误差可能会导致大的预测损失。显然,将低秩近似从训练中分离出来并不是最佳的。与以往的工作,本文集成了低秩近似和正则化的训练过程。我们提出了训练秩修剪(TRP),它交替之间的低秩近似和训练。TRP保持了原始网络的容量,同时在训练期间施加低秩约束。通过随机次梯度下降优化的核正则化被用来进一步提升TRP中的低秩。用TRP训练的网络本质上具有低秩结构,并且以可忽略的性能损失进行近似,从而消除了低秩近似后的微调。该方法在CIFAR-10和ImageNet上进行了综合评估,优于以前使用低秩近似的压缩方法。我们的代码可在https://github.com/yuhuixu1993/Trained-Rank-Pruning上获得。

英文摘要:

To accelerate DNNs inference, low-rank approximation has been widely adopted because of its solid theoretical rationale and efficient implementations. Several previous works attempted to directly approximate a pre-trained model by low-rank decomposition; however, small approximation errors in parameters can ripple over a large prediction loss. Apparently, it is not optimal to separate low-rank approximation from training. Unlike previous works, this paper integrates low rank approximation and regularization into the training process. We propose Trained Rank Pruning (TRP), which alternates between low rank approximation and training. TRP maintains the capacity of the original network while imposing low-rank constraints during training. A nuclear regularization optimized by stochastic sub-gradient descent is utilized to further promote low rank in TRP. Networks trained with TRP has a low-rank structure in nature, and is approximated with negligible performance loss, thus eliminating fine-tuning after low rank approximation. The proposed method is comprehensively evaluated on CIFAR-10 and ImageNet, outperforming previous compression counterparts using low rank approximation. Our code is available at: https://github.com/yuhuixu1993/Trained-Rank-Pruning.

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。

↑