Research &
Scholarship

Leading the advancement of MURDOC into a full mixed-initiative framework enabling bidirectional human-AI collaboration through ROI selection, adjustable sensitivity, and user feedback mechanisms. Developed in response to findings that visual explanations alone improved basic understanding but did not build sufficient trust — particularly among domain experts seeking greater control. MICA integrates domain expertise to support more adaptive, trustworthy AI in operational settings.

University of Oklahoma · DIV-Lab ↗ · Oct 2024–Mar 2026 · Advisor: Dr. G.J. Quadri · Submitted to ACM TOCHI as “Interactive Features and Trust in AI-Assisted Camouflaged Object Detection”

Developed a specialized machine learning model using Low-Rank Adaptation (LoRA) fine-tuning to generate photorealistic camouflaged objects across multiple biological and non-biological categories. Enables controlled variation of camouflage parameters for experimental testing and creation of synthetic datasets to enhance detection model training and evaluation — directly addressing the data scarcity challenge in camouflage research. Submitted to IEEE Computer Graphics and Applications.

University of Oklahoma · Gallogly College of Engineering · Dec 2024–Mar 2026 · Submitted to IEEE Computer Graphics and Applications as “Multi-Architecture Evaluation of Stable Diffusion with HSV–Gestalt Perceptual Loss”

An interactive offline visualization tool integrating Informative AI (IAI) and eXplainable AI (XAI) techniques for camouflage detection. Implements a three-level decision hierarchy using ResNet50 and EfficientDet-D7 to identify object presence, locate weak camouflage regions, and analyze object parts that break concealment. Features Grad-CAM attention visualization, user-controlled image preprocessing, and the TOAST scale for trust measurement. Source code available on GitHub.

IEEE DASC 2024 · Feb 2023–Oct 2024 · github.com/enjelika/MURDOC_2024

Explored GIF and video collections to understand temporal motion patterns using Porter-Duff blend modes (SourceOver, XOR, DestinationOver) and mixed-abstraction visual encoding. Applied to camouflaged wildlife including butterflies, grasshoppers, and geckos, producing composite foreground/background visualizations that make subtle motion perceptible. Directly informed the IEEE VIS 2024 poster and Springer Nature publication.

University of Oklahoma · Gallogly College of Engineering · Jan 2024–Sep 2024

Addressed the challenge of measuring camouflage effectiveness in real-world conditions — beyond controlled lab environments. Developed robust methodologies to quantify, characterize, and optimize camouflage strategies using multidimensional segmentation techniques incorporating color, geometric structure, and validated metrics. Provided controllable synthetic inputs for testing hypotheses about camouflage effectiveness across diverse environments.

University of Oklahoma · Gallogly College of Engineering · Jul 2023–Jan 2026

Bridged the gap between camouflaged object detection/segmentation and self-explainable AI (S-XAI) by leveraging the Self-Explaining Decision Architecture (SEDA) with RankNet for parallel concealed object localization and camouflage ranking. Enables identification of object parts that compromise concealment and deepens understanding of visual cues that subvert camouflage — the direct predecessor to MURDOC and the MICA framework.

United States Air Force · Mar 2023–Jun 2023 · Foundational work informing the DASC 2023 paper

Developed an XGBoost model achieving 96.9% AUC (95% CI: 0.966–0.972) for predicting overall survival in diffuse large B-cell lymphoma patients, outperforming traditional Ann Arbor staging and other ML approaches. Engineered a data preprocessing pipeline for 64,912 SEER patient records with feature extraction, one-hot encoding, and missing value handling. Achieved 92.9% sensitivity, 87.8% specificity, and 91.2% accuracy.

University of Oklahoma · Gallogly College of Engineering · Mar 2022–Nov 2024 · Co-author · Published: ASH Annual Meeting / Blood 2023

Collaborated with Memorial Sloan Kettering Cancer Center to design and develop an interactive hierarchical visualization application for exploring 1,082 TCGA mRNA expression z-score breast cancer records from the cBioPortal. Applied statistical techniques to enable medical researchers to discover new genomic patterns for reevaluating categorization and improving treatment outcomes.

University of Oklahoma · Gallogly College of Engineering · May 2022–Jan 2023

An original theoretical concept extending dissertation findings on time-pressure and AI over-reliance into a structural critique of institutional AI adoption. The central insight: the conditions that drive fastest AI adoption — crisis, urgency, institutional stress — are precisely those least conducive to careful evaluation and trust calibration.

Thought Leadership · Conference Pipeline · 2026