Corsework
sta4321
Posts by Collection
courses
portfolio
Research Project
Active research project — details coming soon.
DataSci-Coder: Fine-Tuned LLM for Data Science
Fine-tuned Qwen2.5-Coder-14B via QLoRA on 10,795 curated DS examples. 93.3% instruction compliance vs 91.4% base. Published on HuggingFace.
E-Commerce Shopper Behavior Analysis
ML pipeline predicting purchase intent from 12,330 sessions using clustering, SMOTE, and ensemble methods achieving 0.93 AUC.
AI Driver Risk Scoring Platform
End-to-end Telematics platform using ML to score driver risk with a real-time Streamlit dashboard achieving 0.98 ROC-AUC.
Theoretical Analysis of CNNs for Automatic Seizure Detection in EEG Signals
Undergraduate Honors Thesis using PyTorch, SMOTE, and Lipschitz bounds to achieve 97% accuracy in seizure detection.
publications
Theoretical Analysis of CNNs for Automatic Seizure Detection in EEG Signals
Published in UCF STARS Honors Undergraduate Theses, 2025
1D CNN with Butterworth filtering achieving 97% accuracy and 0.99 AUC on EEG time-series. Formally proves Lipschitz stability bounds (L = 24.72). Advised by Dr. Chudamani Poudyal (SDMSS, UCF).
Recommended citation: Small, J. T. (2025). "Theoretical Analysis of CNNs for Automatic Seizure Detection in EEG Signals." UCF STARS Honors Undergraduate Theses, No. 462.
Download Paper
talks
Theoretical Analysis of CNNs for Automatic Seizure Detection in EEG Signals
Published:
Poster presentation of honors undergraduate thesis research. Demonstrated the 1D CNN architecture for EEG seizure detection, Butterworth filtering pipeline, Lipschitz stability analysis, and frequency domain interpretation of learned features (22 Hz beta-wave band). Results: 97% accuracy, 0.99 AUC on the University of Bonn EEG dataset.
Theoretical Analysis of CNNs for Automatic Seizure Detection in EEG Signals
Published:
Poster presentation of honors undergraduate thesis research at the annual UCF Student Scholar Symposium. Presented CNN-based EEG seizure detection with formal Lipschitz stability bounds (L = 24.72), frequency domain analysis, and results: 97% accuracy, 0.99 AUC.