Olaniyi Olanrewaju Moluga “Larry”

I contribute to the DESIRE project, which explores how deliberative quality and hostility manifest in online political discussions.

My research centers on scaling multi-task Natural Language Processing (NLP) architectures to model high-dimensional discourse metrics:

Hierarchical Architecture: Developing Progressive Layered Extraction (PLE) and Mixture-of-Experts (MoE) heads on top of pre-trained transformer backbones (DeBERTa-v3-large) to separate shared language semantics from domain-specific (Hostility vs. Deliberation) concepts.

High-Dimensional Regression: Benchmarking model performance across 53 distinct continuous discourse indicators.

Uncertainty Optimization: Implementing uncertainty-weighted loss mechanisms and dynamic batching techniques to handle annotator variance and zero-inflation in complex social media datasets.