About Me
Background
I'm a PhD candidate in Computer Science at the University of Dayton, specializing in adversarial robustness enhancement learning and deep learning for reinforcement learning. My research focuses on developing robust machine learning systems that can withstand adversarial attacks while maintaining high performance.
At a glance
- • RL adversarial defenses (Gym: LunarLander, BipedalWalker)
- • Deep learning for time series (LSTM)
- • Recommender systems (KNN/SVD)
- • Computer vision (OpenCV, dlib)
- 🎓 Education: Pursuing a PhD in Computer Science (GPA: 3.95) at the University of Dayton, with a completed Master's in Computer Science (GPA: 4.00).
Skills & Technologies
Experience
University of Dayton
- •Developed an entropy-driven adversarial defense framework in RL environments (Gym Lunar Lander, Bipedal Walker), achieving 95% accuracy and outperforming KL divergence and joint entropy baselines.
- •Simulated adversarial attacks in RL pipelines to evaluate robustness and improve policy behavior.
- •Facial-recognition attendance system in Python/OpenCV/dlib (~99.39% accuracy) for online sessions.
Walmart
- •Researched and implemented GS1 barcode standards to address 10+ case exceptions (oversized, perishable, hazardous items, etc.) at distribution centers.
- •Developed a Python three-phase function for structure validation, checksum generation, and barcode encoding, optimizing scanning processes.
- •Compiled and executed BigQuery SQL searches to collect 100+ diverse test inputs covering barcode types, product categories, and exceptions.
- •Queried and analyzed 100+ test cases in SQL, achieving 100% accuracy compared to legacy logic
Projects
Entropy-driven feature selection for adversarial robustness. 94-95% accuracy across Gym (LunarLander/BipedalWalker). Outperformed KL Divergence and Joint Entropy at detecting adversarial noise.
Deep learning for time-series prediction. Advanced preprocessing, tuning, and regularization (LR scheduling, early stopping). Benchmarked vs MLP/CNN/CNN-LSTM; LSTM captured long-range temporal patterns.
Full-stack recommender (MovieLens 1M). KNN & SVD with Surprise; Streamlit UI with real-time recs. RMSE ~0.87, Precision@10 ~0.81; persistent pipelines & evaluation dashboards.
Classical search & RL policies in a custom environment. DFS, BFS, UCS, and A* (Manhattan). Value iteration, policy iteration, and Q-learning (ε-greedy) with trajectory visualizations.
Publications
Imposter Injection: Learning to Select Features in Reinforcement Learning
Mouse Brain Cell Segmentation in Fluorescence Microscopy Images
Virtual Yoga Instructor with Real-Time Feedback
Predicting an Optimal Medication/Prescription Regimen for Patient Discordant Chronic Comorbidities Using Multi-Output Models
