Multiple differential cryptanalysis
Differential cryptanalysis that exploits many input and output differences at once instead of a single characteristic.
A single differential may be too weak to show up over many rounds, but dozens or hundreds of weak ones can add up to a clear signal. In 2011 Céline Blondeau and Benoît Gérard gave the technique a proper statistical framework and applied it to reduced-round versions of the lightweight cipher PRESENT.
The framework subsumes earlier special cases: truncated differentials are a kind of multiple differential with some bits left unspecified. The difficulty mirrors the linear counterpart — you have to account correctly for how the differentials depend on each other, or the data estimate comes out far too optimistic. Designers of new ciphers therefore have to show resistance to the whole cluster, not just the best single path.