Thoughts on data science, statistics and machine learning.
Book Review: The Great Arc - John Keay
This book invokes two very different reactions in me. The primary reaction is jubilant, almost romantic. The second is gloomy. Imagine you’re watching Oppenheimer: spirits rising until the point the bombs actually drop, after which you feel guilty about having felt good in the first place. The Great Trigonometric Survey was completed over the duration of a better part of a century, across three generations of mathematicians, physicists and surveyors (they were called compasswallahs - I finally see where Rohit Gupta gets his pseudonym), and at the cost of thousands of lives.
Book Review: A Room of One's Own - Virginia Woolf
I went into this essay expecting Virginia Woolf had written about what the eponymous room is like - its design and contents. But she deals with a more fundamental issue - that one needs a room of one’s own. The essays are a fine piece of scholarship. I’d never have thought that Woolf’s characteristic device, the “stream of consciousness” could be used to not only as a writing technique, but also as a powerful pedagogical technique.
Open World Games and the Myth of Sisyphus
To the memory of Kevin Conroy. There was only ever one true Batman. You have been playing for months. Slowly and steadily, you have harvested every collectable - making yourself stronger and stronger until you can kill the toughest enemies. Every enemy defeated, every monster slain. No side quest worth doing remains. Those not worth doing are also done because you are a completionist (which is a dignified way of saying that you have no life).
Effective Train/Test Stratification for Object Detection
TL;DR: Here’s a talk based on this post: There’s an unavoidable, inherent difficulty in fine-tuning deep neural networks, which stems from the lack of training data. It would seem ridiculous to a layperson that a pretrained vision model (containing millions of parameters, trained on millions of images) could learn to solve highly specific problems. Therefore, when fine-tuned models do perform well, they seem all the more miraculous. But on the other hand, we also know that it is easier to move from the general to the specific, than the reverse.