Tag
Singular Learning Theory (SLT) uses algebraic geometry to explain why neural networks generalize well despite their degeneracies, introducing the real log canonical threshold (RLCT) as a measure of model complexity.