Scholarship & Research
Permanent URI for this communityhttps://scholarworks.montana.edu/handle/1/1
Browse
12 results
Search Results
Item An adaptive genetic algorithm for fitting DeGroot opinion diffusion models on social networks(Montana State University - Bozeman, College of Letters & Science, 2022) Johnson, Kara Layne; Chairperson, Graduate Committee: John J. BorkowskiWhile a variety of options are available for modeling opinion diffusion--the process through which opinions change and spread through a social network--current methods focus on modeling the process on online social networks where large quantities of opinion data are readily available. For in-person networks, where data are more difficult to collect, models that predict the opinions of the individuals in the network require that the structure of social influence--who is influenced by whom and to what degree--is specified by the researcher instead of informed by data. In order to fit data-driven opinion diffusion models on small networks with limited data, we developed a genetic algorithm for fitting the DeGroot opinion diffusion model. We detail the algorithm and present simulation studies to assess the algorithm's performance. We find the algorithm is able to recover model parameters across a variety of network and data set conditions, it continues to perform well under the assumption violations expected in practical applications, and the algorithm performance is robust to most choices of hyperparameters. Finally, we present an analysis of data from the study that motivated the methodological development.Item Calibration and characterization of a VNIR hyperspectral imager for produce monitoring(Montana State University - Bozeman, College of Engineering, 2020) Logan, Riley Donovan; Chairperson, Graduate Committee: Joseph A. Shaw; Joseph A. Shaw was a co-author of the article, 'Measuring the polarization response of a VNIR hyperspectral imager' in the journal 'SPIE proceedings' which is contained within this thesis.; Bryan Scherrer, Jacob Senecal, Neil S. Walton, Amy Peerlinck, John W. Sheppard, and Joseph A. Shaw were co-authors of the article, 'Hyperspectral imaging and machine learning for monitoring produce ripeness' in the journal 'SPIE proceedings' which is contained within this thesis.Hyperspectral imaging is a powerful remote sensing tool capable of capturing rich spectral and spatial information. Although the origins of hyperspectral imaging are in terrestrial remote sensing, new applications are emerging rapidly. Owing to its non-destructive nature, hyperspectral imaging has become a useful tool for monitoring produce ripeness. This paper describes the process of characterizing and calibrating a visible near-infrared (VNIR) hyperspectral imager for obtaining accurate images of produce to be used in machine learning algorithms for analysis. In this work, many calibrations and characterization are outlined, including: a radiance calibration, the process of calculating reflectance, pixel uniformity and image stability testing, spectral characterization, illumination source analysis, and measurement of the polarization response. The images obtained by the calibrated hyperspectral imager were converted to reflectance across a spectral range of 387.12 nm to 1023.5 nm, with a spectral resolution of 2.12 nm. A convolutional neural network was used to perform age classification for Yukon Gold potatoes, bananas, and green peppers. Additionally, a genetic algorithm was used to determine the wavelengths carrying the most useful information for age classification. Experiments were run using red green blue (RGB) images, full-spectrum hyperspectral images, and the wavelengths selected by the genetic algorithm feature selection method. Preliminary data from these analyses show promising results at accurately classifying produce age. The genetic algorithm feature selection method is being used to develop a low-cost multispectral imager for use in monitoring produce in grocery stores.Item Utilizing distributions of variable influence for feature selection in hyperspectral images(Montana State University - Bozeman, College of Engineering, 2019) Walton, Neil Stewart; Chairperson, Graduate Committee: John SheppardOptical sensing has been applied as an important tool in many different domains. Specifically, hyperspectral imaging has enjoyed success in a variety of tasks ranging from plant species classification to ripeness evaluation in produce. Although effective, hyperspectral imaging can be prohibitively expensive to deploy at scale. In the first half of this thesis, we develop a method to assist in designing a low-cost multispectral imager for produce monitoring by using a genetic algorithm (GA) that simultaneously selects a subset of informative wavelengths and identifies effective filter bandwidths for such an imager. Instead of selecting the single fittest member of the final population as our solution, we fit a univariate Gaussian mixture model to a histogram of the overall GA population, selecting the wavelengths associated with the peaks of the distributions as our solution. By evaluating the entire population, rather than a single solution, we are also able to specify filter bandwidths by calculating the standard deviations of the Gaussian distributions and computing the full-width at half-maximum values. In our experiments, we find that this novel histogram-based method for feature selection is effective when compared to both the standard GA and partial least squares discriminant analysis. In the second half of this thesis, we investigate how common feature selection frameworks such as feature ranking, forward selection, and backward elimination break down when faced with the multicollinearity present in hyperspectral data. We then propose two novel algorithms, Variable Importance for Distribution-based Feature Selection (VI-DFS) and Layer-wise Relevance Propagation for Distribution-based Feature Selection (LRP-DFS), that make use of variable importance and feature relevance, respectively. Both methods operate by fitting Gaussian mixture models to the plots of their respective scores over the input wavelengths and select the wavelengths associated with the peaks of each Gaussian component. In our experiments, we find that both novel methods outperform variable ranking, forward selection, and backward elimination and are competitive with the genetic algorithm over all datasets considered.Item Improving a precision agriculture on-farm experimentation workflow through machine learning(Montana State University - Bozeman, College of Engineering, 2019) Peerlinck, Amy; Chairperson, Graduate Committee: John SheppardReducing environmental impact while simultaneously improving net return of crops is one of the key goals of Precision Agriculture (PA). To this end, an on-farm experimentation workflow was created that focuses on reducing the applied nitrogen (N) rate through variable rate application (VRA). The first step in the process, after gathering initial data from the farmers, creates experimental randomly stratified N prescription maps. One of the main concerns that arises for farmers within these maps is the large jumps in N rate between consecutive cells. To this end we successfully develop and apply a Genetic Algorithm to minimize rate jumps while maintaining stratification across yield and protein bins. The ultimate goal of the on-farm experiments is to determine the final N rate to be applied. This is accomplished by optimizing a net return function based on yield and protein prediction. Currently, these predictions are often done with simple linear and non-linear regression models. Our work introduces six different machine learning (ML) models for improving this task: a single layer feed-forward neural network (FFNN), a stacked auto-encoder (SAE), three different AdaBoost ensembles, and a bagging ensemble. The AdaBoost and bagging methods each use a single layer FFNN as its weak model. Furthermore, a simple spatial analysis is performed to create spatial data sets, to better represent the inherent spatial nature of the field data. These methods are applied to four actual fields' yield and protein data. The spatial data is shown to improve accuracy for most yield models. It does not perform as well on the protein data, possibly due to the small size of these data sets, resulting in a sparse data set and potential overfitting of the models. When comparing the predictive models, the deep network performed better than the shallow network, and the ensemble methods outperformed both the SAE and a single FFNN. Out of the four different ensemble methods, bagging had the most consistent performance across the yield and protein data sets. Overall, spatial bagging using FFNNs as the weak learner has the best performance for both yield and protein prediction.Item Robust and optimal design strategies for nonlinear models using genetic algorithms(Montana State University - Bozeman, College of Letters & Science, 2014) Akapame, Sydney Kwasi; Chairperson, Graduate Committee: John J. BorkowskiExperimental design pervades all areas of scientific inquiry. The central idea behind many designed experiments is to improve or optimize inference about the quantities of interest in a statistical model. Thus, the strengths of any inferences made will be dependent on the choice of the experimental design and the statistical model. Any design that optimizes some statistical property will be referred to as an optimal design. In the main, most of the literature has focused on optimal designs for linear models such as low-order polynomials. While such models are widely applicable in some areas, they are unsuitable as approximations for data generated by systems or mechanisms that are nonlinear. Unlike linear models, nonlinear models have the unique property that the optimal designs for estimating their model parameters depend on the unknown model parameters. This dissertation addresses several strategies to choose experimental designs in nonlinear model situations. Attempts at solving the nonlinear design problem have included locally optimal designs, sequential designs and Bayesian optimal designs. Locally optimal designs are optimal designs conditional on a particular guess of the parameter vector. Although these designs are useful in certain situations, they tend to be sub-optimal if the guess is far from the truth. Sequential designs are based on repeated experimentation and tend to be expensive. Bayesian optimal designs generalize locally optimal designs by averaging a design optimality criterion over a prior distribution, but tend to be sensitive to the choice of prior distribution. More importantly, in cases where multiple priors are elicited from a group of experts, designs are required that are robust to the class (or range) of prior distributions. New robust design criteria to address the issue of robustness are proposed in this dissertation. In addition, designs based on axiomatic methods for pooling prior distributions are obtained. Efficient algorithms for generating designs are also required. In this research, genetic algorithms (GAs) are used for design generation in the MATLAB® computing environment. A new genetic operator suited to the design problem is developed and used. Existing designs in the published literature are improved using GAs.Item An empirical study of parallel genetic algorithms for the traveling salesman and job-shop scheduling problems(Montana State University - Bozeman, College of Engineering, 2001) Helzer, David MichaelItem Two heuristic stochastic algorithms applied to a temporally constrained scheduling program(Montana State University - Bozeman, College of Engineering, 2001) Furois, Scott AlanItem Hybridizing statistics with genetic algorithms(Montana State University - Bozeman, College of Letters & Science, 1995) Pamplin, Trenton L.Item An enumerative approach to computing cut sets in metabolic networks(Montana State University - Bozeman, College of Engineering, 2013) Salinas, Daniel; Chairperson, Graduate Committee: Brendan MumeyThe productivity of organisms used in biotechnology may be enhanced when certain parts of their metabolism are rendered inaccessible. This can be achieved with genetic modifications, but current techniques set a practical limit on number of modifications that can be applied. Taking advantage of this limit, we implement a brute force algorithm that can compute cut sets for any set of metabolites and reactions that is shown to perform better than alternative approaches. Also, an attempt is made to approximate a binary linear program with a quadratic program; this approximation is meant to be used when refining the growth model of organisms used in flux balance analysis. The approximation is shown to be less efficient that the original program. Finally, extensions to the brute force algorithm are proposed.Item Project task : member assignment using design structure matrix and genetic algorithm in concurrent engineering project management(Montana State University - Bozeman, College of Engineering, 2005) Mazur, Lukasz Maciej; Chairperson, Graduate Committee: Shi-Jie (Gary) ChenIn concurrent engineering, project tasks generally require the establishment of multifunctional teams to simultaneously consider various activities throughout the entire product life cycle. Team members from different functional departments of the company interact in every phase of development activities to design products and processes concurrently. This concurrent strategy increases the complexity of product development and design processes and makes teams difficult to organize. Without effective task coordination and team organization, the lack of communication and cooperation among team members in a large group of tasks could seriously delay the project completion. This research provides an integrated solution to overcome these difficulties. This research aims to model both project tasks and team members for the task-member assignments. To accomplish this, we develop an integrated framework that includes three major components: a project task model, a team member model and a task-member assignment model. The project task model optimizes the complex task structure using a Genetic Algorithm (GA), while Design Structure Matrix (DSM) identifies the three major project task types: independent, dependent, and interdependent. The team member model provides a quantitative representation for three important team member characteristics, namely functional knowledge, teamwork capability and working relationship. Analytic Hierarchy Process (AHP) and personality profiling using Myers-Briggs Type Indicator (MBTI) are used to obtain ratings of team member characteristics. According to the project task structure, quantified team member characteristics, and each member's workload schedule, the task-member assignment model accomplishes the ultimate goal of this research - assigning the right team members to the right tasks at the right time. The effectiveness of the developed methodology is demonstrated by an illustrative example.