Mathematical Methods of AI

A grab-bag of mathematically grounded AI work: explainable AI, recommender systems, Kalman filters and LDA topic modeling, with a focus on interpretability.

Source on GitHub

For this sample the model favors Neutral (0.68) over Emotional (0.28); the table lists the features behind that call.Open full size: For this sample the model favors

A portfolio of small, mathematically rigorous AI projects: model-agnostic explainability methods, a Kalman filter implementation, LDA-based topic discovery on a text corpus, and a recommender experiment. Each piece prioritizes interpretability over raw accuracy.

Figures

Distributions of the two top random-forest features, Mental against Emotional.Open full size: Distributions of the two top random-forest

The ten hardest items to predict, comparing KNN, baseline and SVD error.Open full size: The ten hardest items to predict

Residuals for X and Y with 30% of observations missing.Open full size: Residuals for X and Y with

LDA topic model: intertopic distance map and the 30 most relevant terms.Open full size: LDA topic model: intertopic distance map

Correspondence

Research, data science work or a question about a project. Email, LinkedIn or the contact form all reach me.

mawangot2000@gmail.com LinkedIn

There is also a contact form, and aresume (PDF).