Doctor of Philosophy (PhD)
Physics & Astronomy
Methodologies from data science and machine learning, both new and old, provide an exciting opportunity to investigate physical systems using extremely expressive statistical modeling techniques. Physical transitions are of particular interest, as they are accompanied by pattern changes in the configurations of the systems. Detecting and characterizing pattern changes in data happens to be a particular strength of statistical modeling in data science, especially with the highly expressive and flexible neural network models that have become increasingly computationally accessible in recent years through performance improvements in both hardware and algorithmic implementations. Conceptually, the machine learning approach can be regarded as one that employing algorithms that eschew explicit instructions in favor of strategies based around pattern extraction and inference driven by statistical analysis and large complex data sets. This allows for the investigation of physical systems using only raw configurational information to make inferences instead of relying on physical information obtained from a priori knowledge of the system. This work focuses on the extraction of useful compressed representations of physical configurations from systems of interest to automate phase classification tasks in addition to the identification of critical points and crossover regions.
Walker, Nicholas, "Identifying Structure Transitions Using Machine Learning Methods" (2020). LSU Doctoral Dissertations. 5311.