Modern machine learning is powerful but often opaque. I make models interpretable and identifiable by grounding them in information theory rather than black-box heuristics. At the center is entropic regression, which is itself a causality-detection approach: it keeps only the variables that genuinely drive a system, yielding sparse, human-readable models whose terms map to real mechanisms instead of spurious correlations. This is exactly where explainability and causal discovery meet, because a model that identifies causal structure is explainable by construction: you can see why a prediction is made and which factors are responsible. I pair this with a family of geometric, entropy-based measures (geometric partition entropy and Boltzmann-Shannon interaction entropy) that quantify information and dependence directly from continuous data, sharpen mutual-information estimation when informative outliers are present, and act as transparent feature-selection and data-quality metrics. Together these tools deliver AI that stays robust on noisy or limited data and remains accountable enough for high-stakes and regulated settings.
AFRL CRADA: Geometric Information Theory and Causation Inference (2023–2025). AFRL VFRP: Spectral Hierarchy Measure using Geometric Partition Entropy (2022).