← Back to Basic Science

HiPPO

2020
Computer Science (theoretical)Machine Learning TheoryFrameworkfoundational

Framework for online function approximation via optimal polynomial projections, giving recurrent models a principled way to compress an arbitrarily long history into a fixed-size state (Gu, Dao, Ermon, Rudra & Ré, "HiPPO: Recurrent Memory with Optimal Polynomial Projections," NeurIPS 2020). The memory theory the entire S4/Mamba line of state space models is built on.

Originators

  • Gu, A.
  • Dao, T.
  • Ermon, S.
  • Rudra, A.
  • Ré, C.

Landmark Paper

W3064840847 ↗
Not retracted (OpenAlex)

Checked 2026-09-19 — interim signal only, see docs/BASIC_ROADMAP.md Phase 10

Connections

  • is component of Linear State-Space Layer (LSSL)
    basis: reasoned

    Appendix B.1 of the Mamba paper: S4/LSSL "focused on efficient convolutional algorithms for DPLR SSMs due to a connection to continuous-time online memorization (HIPPO (Gu, Dao, et al. 2020))" -- HiPPO's memory theory is named as the specific basis LSSL builds its layer on.