Electrical and Computer Engineering ETDs

Publication Date

Spring 5-16-2026

Abstract

This dissertation demonstrates the applications and comparative analyses of machine learning methods in ultrafast laser control. By learning the relationship between the system’s input parameters and output pulse characteristics, the performance of a laser can be significantly improved. In this work, the results are presented in two stages by utilizing data from the femtosecond laser system. The first stage concerns two neural networks, named NN1 (fitrnet) and NN2 (feedforwardnet). The second stage, which extended with five different models, namely the linear regression (fitlm), the support vector machine (SVM), the Gaussian process regression (GPR), the boosted tree (fitrensemble), and LASSO (fitrlinear), was implemented. The detailed results of a comparison study for the models are being presented to demonstrate the performance by calculating MAE, MSE, and R2. It has been observed that GPR showed the best results among all, with an MAE of 0.0205 while holding MSE and R2 values of 0.0015 and 0.7678. These models are used to identify correlations between DAZZLER inputs and FROG outputs, which can guide optimization of the laser pulse system in future. This can lead to the development of such a controller that can generate better laser pulses and ultimately better electron pulses on the laser-driven control system.

Project Sponsors

CBB, DOE

Document Type

Dissertation

Language

English

Degree Name

Computer Engineering

Level of Degree

Doctoral

Department Name

Electrical and Computer Engineering

First Committee Member (Chair)

Ganesh Balakrishnan

Second Committee Member

Michael Devetsikiotis

Third Committee Member

Sang M Han

Fourth Committee Member

Payman Zarkesh Ha

Download

Share

COinS