Critical feature space for predicting the glass forming ability of metallic alloys revealed by machine learning

Progress and Perspective of Metallic Glasses for Energy Conversion and Storage

“…When new electrochemical systems are involved, it is inconvenient to employ the previous calculation model. [201] Owing to the sophisticated requirements of the basic material physical and chemical properties, the new problems are indeed hard to be solved by computational simulation. [202] Therefore, the big database support, powerful data management and sharing platform, as well as powerful machine learning are urgently needed to find patterns in high-dimensional data.…”

Recent development of chemically complex metallic glasses: from accelerated compositional design, additive manufacturing to novel applications

“…With the established dataset, people have applied a variety of ML algorithms to study MGs with a good glass-forming likelihood [17, 71, 74-76, 78, 79, 85-87] (see figures 3(b), (c) and (e)) or a good GFA [17,[74][75][76][87][88][89][90][91][92] (see figures 3(b), (d) and (f)). These mainly include support vector machine (SVM) [17,74,78,79,92], random forest (RF) [17,66,75,76,90], Gaussian process regression (GPR) [17,74] and artificial neural network (ANN) [17,71,79,87,89]. However, to effectively train an ML model, one has to design proper 'fingerprints' or descriptors for their data (data featurization), as shown in figure 3(b).…”

“…However, to effectively train an ML model, one has to design proper 'fingerprints' or descriptors for their data (data featurization), as shown in figure 3(b). At the present time, data featurization is mostly based on the chemical composition of alloys [78,87,91] or by translating a chemical composition into empirical parameters guided by the aforementioned empirical rules [66, 71, 74-76, 88-90, 92], such as mean atomic size [74,75,79,87], atomic size difference [71,74,79,90], mean atomic volume [74,90], mixing enthalpy [74,75,79,87], ideal mixing entropy [71,74,75,79,87], mean electronegativity [75,79,87,90], electronegativity difference [71,79], valence electron concentration [71,75,79,87] and calculated density [71,75,87]. If one considers all individual and collective attributes of constituent elements, the number of the data descriptors designed based on the empirical rules could reach 186 [76], which suggests the intrinsic complexity of the ML based design of BMGs.…”

A critical review of the machine learning guided design of metallic glasses for superior glass-forming ability

“…Although one may easily retrieve GFA data from prior works, data pre-screening and/or data transformation in order to rule out low-fidelity data usually needs to be performed. While this process of data pre-processing can be vital for ML modeling, particularly for a limited data size, we note that it has often been neglected in previous studies [14,15,18,42] . As noted by Liu et al [24] and Zhou et al [31] , a GFA dataset built from successful experiments can be significantly biased if it only includes the data for good glassforming alloys, thereby potentially compromising the efficiency of either classification or regression ML models.…”

“…Given that the design of MGs always has a target property, supervised learning, which has been well developed to seek the correlation between input descriptors and output labels, is the most preferred approach in ML modeling for MG design. To date, researchers have already built different supervised learning-based ML models to address various problems related to MG design, such as the GFA of alloys [8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27] , GFA-related characteristic temperatures [13,14,19,20,23,25,28,29] and various other mechanical and magnetic properties [19,22,30] . With the hitherto reported data, one can design data descriptors based on compositional information, empirical parameters or physical theories.…”