Ongoing investigations into machine learning, digital twins, and the intersection of physical conservation and digital state-space.
2026AIES 2026, under review
Aesthetics Is Not a Mirror: Recursive Aesthetic Reinforcement (under review)
A held-out falsification study of representational bias in widely used text-to-image diffusion models, showing strong statistical convergence toward a narrow occupational aesthetic across thousands of generations.
Technical Focus
Pearson r = 0.873 between LAION-Aesthetics V2 score and the model's outputs; a 200,000-trial Monte Carlo joint null test (p < 1e-5); FairFace cross-classifier replication; a pre-registered held-out falsification. First-author poster at the Midwest Graduate Symposium 2026.
2026Dr. Aniruddha Ray Lab, UToledo
DP-4DGS: Dynamic Priors for 4D Gaussian Splatting
Working draft. Coupling 4D Gaussian splat scene reconstruction with a physics-informed prior to improve dynamic-scene fidelity under sparse observation.
Technical Focus
First-author preliminary results submitted to the Midwest Graduate Symposium 2026. Manuscript in preparation for a graphics and vision venue.
2025IEEE QPAIN 2025, published
Bengali Fake News Detection: A Multi-Layered LSTM Ensemble Approach
A multi-layer LSTM ensemble for detecting misinformation in Bengali-language sources, evaluated at the corpus level.
Technical Focus
An ensemble of multi-layered LSTMs trained on a curated dataset of verified Bengali news, with weighted prediction merging, reaching 82.43% accuracy.
Context-Aware Dialectal Arabic Machine Translation with Interactive Region and Register Selection
A context-aware, steerable framework for dialectal Arabic machine translation that models regional and sociolinguistic variation explicitly.
Technical Focus
A rule-based data augmentation pipeline expands a 3,000-sentence seed corpus to 57,000 parallel sentences across eight regional varieties; an mT5-base model is fine-tuned on lightweight metadata tags for controllable region and register.