Designed and implemented GenCyberSynth, a unified experimental framework comparing seven probabilistic generative models for 40×40 grayscale malware images (GAN variants, VAEs, diffusion, and flow-based models). Built a standardized PyTorch pipeline on UND’s Talon HPC cluster with consistent preprocessing, fixed seeds, and a shared metric suite (FID/cFID, JS/KL, diversity, minority-class F1, and calibration using a frozen CNN classifier). Analyzed trade-offs between sample quality, class balance, and downstream detection performance to inform my PhD research on generative APT simulation and reinforcement-learning–based cloud defense.

Collaborated on a Bayesian neural network–based CSI feedback framework (BCsiNet) for massive MIMO systems. Helped design the encoder–decoder that compresses and reconstructs CSI while using uncertainty estimation to detect interference on the feedback link. Implemented the simulation pipeline with the COST 2100 channel model and benchmarked against CsiNet/CsiNet+, showing that BCsiNet preserves reconstruction quality while reliably flagging corrupted feedback, improving robustness in non-ideal wireless environments. Published in Proceedings of the International Symposium on Intelligent Computing and Networking 2025 (ISICN 2025)

Led research on using masked autoencoders to recover missing values in complex polymer datasets. Proposed a custom masking strategy tailored to chemical data structure. Achieved high accuracy in recovery and robustness compared to baseline imputation methods. Presented findings at IEEE CARS 2024.

Developed and presented a case study applying explainable AI (XAI) techniques to predict cloud point temperature in polymer-related data. Built a predictive modeling workflow, analyzed key feature contributions using interpretable model explanations, and validated results for clarity and scientific relevance. Presented findings as a poster at the 2024 Red River Valley American Chemical Society (RRV ACS) Conference in Bemidji, MN, communicating the methodological approach, results, and implications for data-driven materials research.