Archives
All the articles I've archived.
2026 2
June 2
-
From 86% to 71%: learning to trust a brain decoder's accuracy
I trained a tiny CNN to read MEG brain signals on Colab. The interesting part wasn't the model — it was how stratified splits, checkpoint hygiene, and cross-validation deflated a flattering 86% into an honest 71% ± 5%, and revealed exactly where the model wins.
-
How I use Google Colab
Notes and recipes from my day-to-day Google Colab workflow — setup, data handling, and package installs.