Debate over engram overfitting under data repetition
A study warns that engram/n-gram methods may significantly amplify overfitting when training data is repeated, but engineer giffmana argues this is just ordinary overfitting, noting the effect naturally appears when epoch boundaries are clear rather than shuffled.
2026-10-11 ~ 2026-10-11 · 2 related posts
- Warning: engram/n-gram methods may markedly amplify overfitting under data repeat — SonglinYang4 · 2026-10-11
- 'Nothing new': researcher says engram overfitting is plain overfitting tied to over-parametrization — teortaxesTex · 2026-10-11