Blog / Machine Learning
Essays on how learning systems work and how they fail. Each piece is grounded in the established literature, references the papers it leans on, and comes with a notebook you can run yourself whenever a claim can be shown rather than asserted.
This one started in the margins of my PhD student's final thesis draft: his avionics models were trained once and evaluated once, but the data they describe spans decades. What catastrophic forgetting is and where the standard story comes from, which remedies I trust and why, how to read the field's numbers with care, a reading path through the literature, and a two-minute experiment you can run on a laptop.