Masters Projects
During my masters degree I worked on many diffrent projects for diffrent classes. Here is a list of some that are of note:
A Study of the Effectiveness of Artificial Neural Networks on Determining the Best Academic Learning Modality for Students Studying Math
Abstract: This project discusses the use of Artificial Neural Networks (ANNs), and its ability to determine suitable learning modalities for students after taking a quiz. Our literature review found that the use of ANNs led to good results for others performing analysis in the same field. The algorithm we designed is able to deduce a student's learning style by answering a few questions. The efficiency that our program displayed leads us to believe that our model can be used in a real time context to give feedback to students about better studying modalities that they can take into consideration after taking a quiz. Overall the model has been effective on training data, and we are hopeful that our research can be used in future work.
Utilizing Particle Swarm Optimization to Create Stronger Convolutional Neural Networks
Project for my Computer vision class, This project was analizing the practicality of utilizing adaptive layers to create better more concise neural networkd.
Abstract: Proposing the development of a particle swarm algorithm to assist in hyperparameter selection. This algorithm tests sets of hyperparameters at a time attempting to narrow down what set of hyperparameters provides the best convolutional neural network. The results from this project led to the assertion that particle swarm algorithms should be utilized for the selection of hyperparameters in larger search spaces than the one in this experiment. While the particle swarm algorithm does lead to significant improvements in algorithms, the time and effectiveness of the algorithm was not worthwhile for the results
Learning to Judge a Dynamic Scoring Function for Lifelong Navigation
Project for my AI Robotics class. This project was looking to develop better navigation patters for robots using lifelong navigation instead of a traditional planners.
Abstract: Autonomous robots navigating complex environments must balance efficiency and safety while adapting to new conditions. Traditional planners like the Dynamic Window Approach (DWA) often exhibit rigid behaviors, limiting performance in dynamic tasks. Lifelong Learning for Navigation (LLfN) improves adaptability by combining classical planning with learned policies, mediated by a scoring function that decides which controller to trust. However, existing scorers rely on fixed heuristics, restricting generalization. We propose a dynamic learned scoring function, D^ψ(s,a), trained on long-term outcomes and integrated with Gradient Episodic Memory (GEM) for continual learning. In experiments across 400 episodes, our approach achieved a 74.25% success rate, reduced collision frequency to 0.072 per episode, and demonstrated improved arbitration between planners (average ratio 0.142 during training, 0.205 in testing). These results support our thesis: a learned dynamic scorer significantly enhances planner arbitration and adaptability compared to heuristic-based LLfN, moving toward practical lifelong navigation in complex environments.
Predicting Air Quality Using Ensemble Learning Techniques
Team project for my Machine Learning class. This study found that the use of ensemble learning models like Random Forest and Gradient Boosting for air quality prediction is promising. Especially due to the noisy and complex high dimensional data of air pollutant concentration.
