Teaching a Random Forest to resist color shortcuts during training
A walkthrough from diagnosing color bias in Random Forests to testing an interpretability-guided intervention during tree growth.
Journal / small projects
Small projects, rough ideas, and the details worth remembering.
A walkthrough from diagnosing color bias in Random Forests to testing an interpretability-guided intervention during tree growth.
The same question, a different Gemma: an Italian artist, dinner that must be pasta, and a suspiciously sauce-related origin story. A small experiment in SAE-based activation steering.