edit / Kaixia
Machine learning has great potential in accelerating material research . Many fields of material science benefit from its applications , However, some challenges remain , Will the field be like the hype surrounding it , Remains to be seen .
The machine age is coming . When we put forward the focus of machine learning in Material Science , We know very well that the algorithm can write a reasonable opening editorial for it . After all , It won't be the first time to write an article , Or in this case , It's not even the first time to write a Book .
You can ask Alexa or Siri, It will use its machine learning algorithm to find some articles for you about the benefits and dangers of artificial intelligence . Based on your past searches and their interest in you , It may continue to speculate whether there is too much hype about the ability of machine learning tools to surpass human beings .
However , without doubt , Machine learning is affecting all areas of Science , Material science is no exception . stay 《Nature Reviews Materials》 Magazine 2021 year 8 month , The first 6 volume The first 8 The focus of this issue and other articles in the magazine , This paper discusses how machine learning promotes material research , What progress can we actually expect , And what researchers should pay attention to to to ensure that their machine learning algorithms work in the way they are designed .
Machine learning and emerging materials intelligent ecosystem
To set the scene , From Argonne National Laboratory 、 Georgia institute of technology （Georgia Institute of Technology） Of scientists in 《Nature Reviews Materials》 The magazine is entitled 《Emerging materials intelligence ecosystems propelled by machine learning》 An overview of . The key components of machine learning in materials science are reviewed , From data acquisition and management protocol to independent experiment strategy . The imminent challenges were discussed , This includes the standardization protocol 、 The need for open sharing of benchmark data sets and machine learning code and data .
Thesis link ：https://www.nature.com/articles/s41578-020-00255-y
Machine learning and material design
Machine learning toolbox has been effectively applied in almost all topics of material science . It is used in the design of photonic devices ,
Researchers in the Department of electrical engineering at Stanford University wrote a paper entitled 《Deep neural networks for the evaluation and design of photonic devices》 This is described in detail in the review of .
In this review , The researchers showed how the deep neural network configured as a discriminant network learns from the training set and runs as a high-speed proxy electromagnetic solver . This paper studies how to learn the geometric features of device distribution in deep generation network , Even configured as a powerful global optimizer . The basic data science concept constructed in the context of photonics is discussed , Including network training process 、 Division of different network categories and architectures , And dimensionality reduction .
Thesis link ：https://www.nature.com/articles/s41578-020-00260-1
Just like Rutgers University （The State University of New Jersey）Adam Gormley And Princeton University Michael Webb In a review 《Machine learning in combinatorial polymer chemistry》 As described in , Machine learning can control the complex structure of Polymers — The ability of functional landscape can be used to design high-performance polymers .
Thesis link ：https://www.nature.com/articles/s41578-021-00282-3
Another example is Brigham Young University （Brigham Young University,BYU）Gus Hart And colleagues at 《Machine learning for alloys》 As discussed in the comments , When applied to alloys , Machine learning has promoted the progress of material optimization from metallic glass to high entropy alloys and structural materials .
This paper summarizes the research status of machine learning driven alloys , The methods and applications in this field are discussed , The theoretical prediction and experimental verification are summarized . It is foreseen that the cooperation between machine learning and alloys will lead to the design of new and improved systems .
Thesis link ：https://www.nature.com/articles/s41578-021-00340-w
Besides , Machine learning accelerates and improves the synthesis of nanoparticles with various properties , University of Toronto Eugenia Kumacheva And his colleagues published a review 《Nanoparticle synthesis assisted by machine learning》. This review discusses the methods that can be used in the synthesis of nanoparticles ML Algorithm , The key methods of collecting large data sets are introduced . Studied ML Guided semiconductor 、 Metal 、 Synthesis of carbon based and polymer nanoparticles , Finally, we discuss ML Current limitations in the development of assisted nanoparticle synthesis 、 Advantages and Prospects .
Thesis link ：https://www.nature.com/articles/s41578-021-00337-5
Machine learning and biomedicine
As in the United States Simons The foundation Flatiron As the researchers at the Institute discussed in their comments , Machine learning has also become a key tool for interrogating complex and large biomedical data sets , Ability to study multicellular complexity and develop personalized therapies .
First , This paper briefly introduces the key single-cell and whole tissue methods that allow the study of tissue specificity , then , Two kinds of methods based on machine learning are introduced , They can be used to analyze 、 Model and interpret the experimental data of these methods . Last , Combined with high resolution 、 Large scale multiomics data sets and interpretable machine learning models , The future possibility of examining multicellular complexity is prospected .
Thesis link ：https://www.nature.com/articles/s41578-021-00339-3
Machine learning and computing sustainability applications
Machine learning can be applied to very different problems , Therefore, machine learning can promote the cross integration between disciplines . In the past decade , The Institute for computational sustainability has developed artificial intelligence and software for various applications of computational sustainability ML Method , These methods are also used to advance material discovery , vice versa .
Scientists from Cornell University and the California Institute of technology published 《Computational sustainability meets materials science》 A review article entitled . The synergy between ecology and materials science applications is described . It emphasizes the interpretability of the model and the combination of previous scientific knowledge to better adjust the model 、 Produce scientific solutions , And make up for the lack of training data .
Thesis link ：https://www.nature.com/articles/s41578-021-00348-2
Machine learning and text mining
A promising application of machine learning is text mining , Extract information from articles and other documents , And integrate it into structured data sets . Researchers at the Polytechnic University of Catalonia published 《Time to kick-start text mining for biomaterials》 Comments entitled , This feat is particularly important for complex and difficult to organize data , For example, biomaterial data . Applying text and data mining tools to biomaterials first needs to solve several specific field challenges . These include the high heterogeneity of data and the multidisciplinary and rapidly developing language used in biomaterial publications .
Thesis link ：https://www.nature.com/articles/s41578-020-0215-z
Machine learning and quantum technology
Another important trend is to promote the ability to experiment 、 An autonomous system that measures the results and makes decisions in the next iteration . This method is particularly advantageous for experiments that require cumbersome manual optimization ： One example is the optimization of superconducting qubits , As Oxford University Natalia Ares stay 《Machine learning as an enabler of qubit scalability》 As discussed in , In this field , Machine learning may be the key factor to realize the scalability of related qubits .
Thesis link ：https://www.nature.com/articles/s41578-021-00321-z
Machine learning and industrial applications
Like other fields of Science , From principle verification demonstration to practical application , Synergies with industrial partners will be crucial . Toyota Institute Muratahan Aykol And colleagues at 《Machine learning for continuous innovation in battery technologies》 I think , This is especially true for the application of machine learning in battery optimization , Under real conditions, the data obtained from the test is essential . The researchers discussed data-driven 、 How can machine learning based methods help battery researchers meet the needs of battery continuous innovation .
Thesis link ：https://www.nature.com/articles/s41578-020-0216-y
To explore the views of an industrial researcher ,《Nature Reviews Materials》 The magazine interviewed then Google Research The engineer Patrick Riley, He said ,「 As the field matures , An important trend is to turn to machine learning as an independent component system , But as a good integrated gear .」
Thesis link ：https://www.nature.com/articles/s41578-021-00349-1
All the promises of machine learning may not be realized . Machine learning is a powerful tool , But let machine learning algorithms match the enthusiasm and creativity of researchers who contribute to material science , It will take a long time . This may be a good thing .
Reference link ：https://www.nature.com/articles/s41578-021-00351-7