Metrics
3 File views/ downloads
26 Record Views
Thesis
Machine Learning Models for Analyzing Working Memory Data
University of West Florida Libraries
Master of Science (MS), University of West Florida
2024
Abstract
In primate neurophysiology, it is consistently observed that during working memory tasks, neural tuning manifests as heightened firing rates, persisting throughout the delay period of the task. This study leverages supervised machine learning methodologies to construct models which decode the activity of collections of single neurons during the delay period of the oculomotor delayed response (ODR) task, achieving a prediction accuracy exceeding 90%. This underscores the robustness and fidelity of the information encoded within the neural activity.Experimental data were collected from the prefrontal and posterior parietal cortices of three male rhesus monkeys, spanning both juvenile and adult stages. Rigorous statistical techniques were used for feature selection on the time series spike data. Additionally, hyperparameter op- timization was conducted to ensure the models’ efficacy, including their ability to generalize to novel data from withheld trials. The results demonstrate that the firing-rate encoding by single neurons is profoundly informative. Furthermore, aggregations of neurons can significantly bolster the accuracy of predictions, even when utilizing relatively simple models such as logistic regres- sion, support vector machines, and shallow artificial neural networks.
Despite notable variations between juvenile and adult monkeys, as well as among individual subjects, the models consistently exhibited strong predictive capabilities across all cases, even for the posterior parietal cortex, which is not traditionally associated with significant working memory function.
These results suggest that machine learning methods can provide additional insights on neural recording data, offering a powerful tool for understanding the neural basis of working memory and potentially other cognitive functions.
Details
- Title
- Machine Learning Models for Analyzing Working Memory Data
- Resource Type
- Thesis
- Contributors
- Shusen Pu (Committee Chair)Maher Amer (Committee Member)Subhash Bagui (Committee Member)Achraf Cohen (Committee Member)Samantha Seals (Committee Member)
- Publisher
- University of West Florida Libraries
- Format
- pdf
- Number of pages
- 81
- Copyright
- © 2024 Jason Gerstenberger
- Identifiers
- 99380569596606600
- Academic Unit
- Mathematics and Statistics
- Language
- English
- Awarding Institution
- University of West Florida; Master of Science (MS)
- Theses and Dissertations
- Master of Science (MS), University of West Florida