kapynResearch

Amortizing Maximum Inner Product Search with Learned Support Functions

Amortized MIPS uses learned neural networks to predict Maximum Inner Product Search solutions. This regression-based approach amortizes the cost of repeated MIPS queries from a known distribution over a fixed key database by leveraging the MIPS value function's relationship with the support function of the key set.

Apple ML Research·Jul 2, 2026

Opening Kapyn…