Log-Quantized Stochastic Computing for Memory and Computation Efficient DNNs
Log-Quantized Stochastic Computing for Memory and Computation Efficient DNNs
复制标题
用于内存和计算效率 DNN 的对数量化随机计算
DOI：
10.1145/3287624.3287714
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Jongeun Lee
中科院分区：
文献类型：
--
作者：
H. Sim;Jongeun Lee
为了提高能量效率，人们提出了许多用于深度神经网络的低位量化方法。其中，对数量化是突出的，表现出可接受的深度学习性能。它还简化了高成本乘法器，并大幅减少了内存占用。同时，为了实现低成本的DNN加速，提出了随机计算方法，而最近提出的随机计算乘法器显著提高了算法的精度和时延，这是随机计算方法的主要缺点。然而，在它们的二进制接口系统中，量化基本上是线性的，与传统的定点二进制一样，其成本仍然比存储所有随机流小得多。我们将对数量化的DNN应用到最先进的SC乘法器中，并研究了它如何受益。我们发现，对数量化输入上的SC乘法更准确，有助于微调过程。此外，我们还利用降低的输入复杂度设计了低成本的SC-DNN加速器。最后，在对数量化有利于数据流的同时，所提出的结构比以前的SC-DNN加速器分别减少了40%和24%的面积和功耗。它的面积×延迟积甚至比基于移位器的加速器还要小。
For energy efficiency, many low-bit quantization methods for deep neural networks (DNNs) have been proposed. Among them, logarithmic quantization is being highlighted showing acceptable deep learning performance. It also simplifies high-cost multipliers as well as reducing memory footprint drastically. Meanwhile, stochastic computing (SC) was proposed for low-cost DNN acceleration and the recently proposed SC multiplier improved the accuracy and latency significantly which are main drawbacks of SC. However, in their binary-interfaced system which yet costs much less than storing all stochastic stream, quantization is basically linear as same as conventional fixed-point binary. We applied logarithmically quantized DNNs to the state-of-the-art SC multiplier and studied how it can benefit. We found that SC multiplication on logarithmically quantized input is more accurate and it can help fine-tuning process. Furthermore, we designed the much low-cost SC-DNN accelerator utilizing the reduced complexity of inputs. Finally, while logarithmic quantization benefits data flow, proposed architecture achieves 40% and 24% less area and power consumption than the previous SC-DNN accelerator. Its area × latency product is smaller even than the shifter based accelerator.