Thomas Lumley • 2/1/2019

Recognising when you don’t know

The article discusses a key limitation in machine learning classifiers, using an xgboost mushroom classification example. It highlights how a model trained on specific genera can dangerously misclassify toxic species from unseen genera, analogous to the human difficulty in admitting uncertainty. It examines the technical challenge of designing models to output 'Don't Know' for out-of-distribution data.

0 comments

#Machine Learning #classification #Xgboost