---
title: "Shun'ichi Amari"
type: "entity"
entity_kind: "person"
status: "hub"
canonical_name: "Shun'ichi Amari"
aliases: ["Shunichi Amari","S. Amari","Amari"]
first_seen: "2026-06-29T00:00:00.000Z"
writer_model: "claude-opus-4-8"
connects_to: ["stochastic gradient descent","associative memory (1972) vs Hopfield","natural gradient","priority vs paternity","Jürgen Schmidhuber (witness)"]
---


Japanese mathematician who is now the vault's uncited prior in *two* founding
neural-network stories — which is what graduated him from a watching stub to a
hub (Cali, 2026-07-15; both criteria met). The 1968 Kyoritsu book was read
directly: it credits Saito's 1967 Kyushu thesis for a stochastic-descent
experiment, real but *without* the "multilayer perceptron" framing that
Schmidhuber's retelling adds — so the SGD-MLP claim gains primary support
while its load-bearing wording stays single-witness and citogenesis-tainted.
Separately, and on firmer footing, his 1972 IEEE associative-memory model has
substantive priority over the Hopfield network and goes uncited in Hopfield
1982. Also originated the natural-gradient method.

## References
- [[myth-amari-first-sgd-mlp]] (the citogenesis case) · [[claim-amari-1968-saito-experiment-primary-read]] · [[claim-amari-1972-associative-memory-precedes-hopfield]] · [[claim-amari-stochastic-descent-and-widrow-hoff-lms-are-sgd-siblings]] · [[claim-amari-1998-natural-gradient-fisher-steepest-descent]] · [[claim-wikipedia-amari-sgd-citogenesis]]
- 1977 neural-field / lateral-inhibition thread: [[claim-amari-1977-neural-field-paper-rooted-in-reaction-diffusion-morphogenesis-literature]] · [[claim-no-amari-1977-neural-field-application-to-morphogenesis-found]] · [[claim-wilson-cowan-and-amari-neural-field-formulations-are-the-same-object]]
- Related hubs: [[entity-john-hopfield]] · [[entity-juergen-schmidhuber]] · [[moc-backpropagation-origins]]