该研究介绍了一种基于深度神经网络的基本新方法，以基于已知的物理模型将函数形式拟合到噪声数据。来自美国橡树林国家实验室的Stephen Jesse领导的团队，提出了一种新的方法，可用来逆向解决问题，可从基于光谱成像数据的最小二乘拟合中提取物理模型参数，并能通过深度学习测定先验参数而增强提取能力。他们将这种方法应用于从压电响应力显微镜数据中提取简谐振子参数，并表明通过结合使用深度神经网络和最小二乘拟合，可以探测比传统方法低一个数量级的信号响应，接近激发信号的热限制。作为模型系统，他们演示了从层状铁电化合物的带激发压电响应力显微镜成像中提取阻尼简谐振子参数。这种使用深度神经网络的方法是通用的，并且在正向和反向情况下都显示出它们作为函数近似器的效用，且在嘈杂的环境中工作良好。该文近期发表于npj Computational Materials 5 25(2019)。
Summary
Physical property from noisy data: Deep neural networks
A fundamentally new method based on deep neural networks, to fit functional forms to noisy data based on a known physical model is introduced. A team led by Stephen Jesse from the Oak Ridge National Laboratory, USA, demonstrated a novel approach for the inverse problem solution and extraction of physical model parameters from spectral-imaging data-based least-squares fitting augmented by deep learning for determination of priors. They apply this method to the extraction of simple harmonic oscillator parameters from piezoresponse force microscopy data, and show that by using a combination of both deep neural networks and least-squares fitting, they can probe signal responses in regimes an order of magnitude lower than with the traditional means, approaching the thermal limit for the excitation signal. As a model system, they demonstrate the extraction of damped simple harmonic oscillator parameters from band-excitation (BE) piezoresponse force microscopy(PFM) imaging of a layered ferroelectric compound. This approach of using deep neural network (DNN) is general and shows their utility as function approximators in both forward and reverse cases and that they work well in noisy environments. This article was recently published in npj Computational Materials 5 25(2019).
原文Abstract及其翻译
Deep neural networks for understanding noisy data applied to physical property extraction in scanning probe microscopy (应用深度神经网络解读扫描探针显微镜中噪声数据的物理特性)
Nikolay Borodinov, Sabine Neumayer, Sergei V. Kalinin, OlgaS. Ovchinnikova, RamaK. Vasudevan & Stephen Jesse
Abstract The rapid development of spectral-imaging methods in scanning probe, electron, and optical microscopy in the last decade have given rise for large multidimensional datasets. In many cases, the reduction of hyperspectral data to the lower-dimension materials-specific parameters is based on functional fitting, where an approximate form of the fitting function is known, but the parameters of the function need to be determined. However, functional fits of noisy data realized via iterative methods, such as least-square gradient descent, often yield spurious results and are very sensitive to initial guesses. Here, we demonstrate an approach for the reduction of the hyperspectral data using a deep neural network approach. A combined deep neural network/least-square approach is shown to improve the effective signal-to-noise ratio of band-excitation piezoresponse force microscopy by more than an order of magnitude, allowing characterization when very small driving signals are used or when a material’s response is weak.
摘要 在过去十年中，扫描探针、电子显微镜和光学显微镜的光谱成像方法发展迅速，导致大型多维数据集的兴起。在许多情况下，将高光谱数据降维到较低维度的材料特性参数，依赖于功能拟合，虽然拟合函数的近似形式是已知的，但函数的参数却是需要人为确定的。然而，通过迭代方法实现噪声数据的功能拟合(如最小二乘梯度下降)，常常出现虚假结果，并且对初始猜测值非常敏感。本研究提出了一种利用深度神经网络方法降维高光谱数据的方法。将深度神经网络与最小二乘法结合使用，可以将带隙-激发压电响应力显微镜的有效信噪比提高一个数量级以上，从而可用非常小的驱动信号即可实现表征，或可用很弱的材料响应即可实现表征。
