Npj Comput. Mater.: 知识的力量—数据与知识的正面较量

海归学者发起的公益学术平台

分享信息，整合资源

交流学术，偶尔风月

Fig. 1 Numerical simulation of water sorption in a porous matrix.

显然，机器学习模型是建立在“数据”训练的基础上，从传统逻辑看，即便“数据”构建中运用了一些“知识”，“数据”的主导地位毋庸质疑。所谓成也数据，败也数据，由数据驱动的机器学习模型通常不具备外延推理能力，原因很简单，魔法的使用是有界限条件的。

Fig. 2 End-to-end differentiable reformulation of the sorption simulation.

如果魔法不能打败魔法，机器学习模型的外推能力如何被拯救？如何要求模型预测一个从未见过的材料性质所对应的结构特征？

Fig. 3 Accuracy of the differentiable simulator.

来自四川大学高分子学院的固体信息学AI实验室（SOFT-AI-Lab）刘晗团队，联合美国加州大学洛杉矶分校与谷歌大脑团队合作者，从机器学习模型视角出发，提出了一个直接由“知识”训练的深度计算框架，让计算模拟与深度学习相互融合，摆脱了对预定义训练数据的依赖，并获得卓越的外延推理能力。

Fig. 4 Training an inverse design generator by the differentiable simulator. 

基于可微分程序与硬件加速，该团体设计一类可微分的材料模拟器，直接端对端接入深度学习生成器模型，通过模拟器的物理知识（而非预定义数据）直接训练生成器，突破了传统反向设计思路的外推精度局限，并采用矩阵处理器实现计算加速。

Fig. 5 Training acceleration by Tensor Processing Unit (TPU) computing.

该研究以多孔材料吸附曲线模拟为例，利用上述深度计算框架，实现了精准预测任意曲线所对应的多孔结构特征，证明以物理知识直接训练生成器将极大提升模型的外延推理能力，同时也突出了矩阵处理器在加速计算模拟应用上的潜力。该研究提出的材料反向设计新架构，将有望推动材料反向设计范式变革，实现高性能新材料的加速开发。

End-to-end differentiability and tensor processing unit computing to accelerate materials’ inverse design 

Han Liu, Yuhan Liu, Kevin Li, Zhangji Zhao, Samuel S. Schoenholz, Ekin D. Cubuk, Puneet Gupta & Mathieu Bauchy

Numerical simulations have revolutionized material design. However, although simulations excel at mapping an input material to its output property, their direct application to inverse design has traditionally been limited by their high computing cost and lack of differentiability. Here, taking the example of the inverse design of a porous matrix featuring targeted sorption isotherm, we introduce a computational inverse design framework that addresses these challenges, by programming differentiable simulation on TensorFlow platform that leverages automated end-to-end differentiation. Thanks to its differentiability, the simulation is used to directly train a deep generative model, which outputs an optimal porous matrix based on an arbitrary input sorption isotherm curve. Importantly, this inverse design pipeline leverages the power of tensor processing units (TPU)—an emerging family of dedicated chips, which, although they are specialized in deep learning, are flexible enough for intensive scientific simulations. This approach holds promise to accelerate inverse materials design.