Blog

Attention as a learned way to decide what context matters for each token.

Bias and variance explain why both too-simple and too-flexible models can fail.

A practical diagnosis map for underfitting, overfitting, model complexity, and generalization.

Bootstrap resampling estimates uncertainty by repeatedly reusing the observed test set.

A model score without uncertainty is easy to read but easy to overtrust.

Correlation is useful evidence, but causal claims require stronger design and stronger assumptions.