Date of Award

Summer 8-15-2018

Author's School

Graduate School of Arts and Sciences

Author's Department


Degree Name

Doctor of Philosophy (PhD)

Degree Type



We introduce a statistical physics inspired supervised machine learning algorithm for classification and regression problems. The algorithm predicts the classification/regression values of new data by combining (via voting) the outputs of these numerous linear expansions in randomly chosen functions. The algorithm has been tested on diverse training data sets of various types and feature space dimensions. It has been shown to consistently exhibit high accuracy and readily allow for optimization of parameters, while simultaneously avoiding pitfalls of existing algorithms such as those associated with class imbalance. We applied this machine learning approach that we term that the "Stochastic Replica Voting Machine" (SRVM) to a binary and a 3-class classification problems in materials science. We employ SRVM to predict candidate compounds capable of forming cubic perovskite (ABX3) structures, double perovskite structures and further classify binary (AB) solids. The results of our binary and ternary classifications compared well to those obtained by the SVM algorithm. In studying double perovskite materials we extended our SRVM analysis to include neural networks and further made comparisons to SVM. At the end we presented neural net and SRVM models to predict the stability of new double perovskite compounds.


English (en)

Chair and Committee

Zohar Nussinov

Committee Members

Alexander Seidel, Li Yang, Rohan Mishra, John Clark,


Permanent URL: 2018-08-15

Available for download on Monday, August 15, 2118

Included in

Physics Commons