No. 09 · Machine Learning
Notes on ML, written to understand them better.
I'm a Machine Learning Engineer at Kensho — S&P Global's AI lab — working on agentic engineering, harness engineering, and the AI-friendly data layer behind both. This is where I work things out in writing: derivations, experiments, and the occasional tangent.
Recent writing
9 essays- SuperBox — what’s really going on with that “free TV” boxA SuperBox is a small box you plug into your TV to watch hundreds of channels for a one-time price. Here’s what’s actually happening behind that “free TV forever” deal.
- Getting Up to Speed with SHAP for Model InterpretabilityValue proposition, intuition, and analytical breakdown of SHAP
- Logistic Regression Deep DiveDerivation of objective and gradient, visualizing a gradient update
- A Quick Primer on KL DivergenceIntuitive introduction to KL divergence, including discussion on its asymmetry
- Latent Variable Models, Expectation Maximization, and Variational InferenceIntroduction to latent variable models, derivation of the ELBO, and the relationship with Expectation Maximization
- Variational Autoencoder TheoryFormulation of the VAE objective, neural network architecture design, optimization, and practical uses
- Blog Post Series: From KL Divergence to Variational Autoencoder in PyTorchLanding page for the blog post series
- Variational Autoencoder Code and ExperimentsDetailed walkthrough of a PyTorch VAE implementation trained on MNIST, including visualizations of data generation, reconstruction, and anomaly detection
- Interesting Details about ROC Curve CalculationsCode-based walkthrough showing how AUC is computed for a simple model trained on a simple dataset