Blog

Building FSTs to recognize poker hands and composing WFSTs for isolated digit speech recognition.

Building skip-gram and CBOW word2vec models with negative sampling, training on the Stanford Sentiment Treebank.

Building logistic regression from scratch in NumPy with gradient checking, feature scaling, and decision boundary visualization on exam data.

Collecting 15K tweets from the Twitter Streaming API, comparing language detection methods, and analyzing geolocation patterns.

Implementing soft-margin linear SVM and polynomial kernel SVM from scratch with gradient descent, grid search over learning rates and regularization.

Visualizing Columbus public transit routes using GTFS data, React, and Leaflet as part of the Smart Columbus initiative.

Implementing binary logistic regression with gradient descent on 784-dimensional image data. Numerically stable gradients and learning rate tuning.

Comparing the Bayes-optimal quadratic boundary to Fisher LDA on 2D Gaussians, then testing VC dimension bounds on a rectangle concept learner.