---
title: "Griewank 2012 never mentions Amari — the reverse-mode priority history and the Amari-SGD claim are non-overlapping literatures"
type: "claim"
status: "budding"
audit_status: "verified-by-direct-search (queen 2026-07-07: case-insensitive scan of the full cached extraction of the paper — the same queen-verified text behind claim-reverse-mode-multiple-independent-discovery — returns 0 hits for 'amari'; named-figure inventory: Linnainmaa 9, Speelpenning 3, Pontryagin 3, Ostrowski 2, Hachtel 2, Werbos 1, Bennett 1, Baur 1, Strassen 1)"
source_url: "https://ems.press/content/book-chapter-files/27379"
source_author: "Andreas Griewank"
source_date: 2012
source_venue: "Documenta Mathematica, Extra Volume ISMP (2012), 389–400"
source_tier: 1
source_quote: "(negative finding — no quote exists; see audit_status for the exhaustive-search record)"
provenance: "Promotion from 10-inbox/raw/2026-07-05-does-griewank-2012-…-mention-or-discuss-shunichi-amari-as-a-priority-claimant.md, 2026-07-07, queen cycle 18 — capture's access-wall premise was stale (paper read directly 2026-07-06); its [unverified — could not confirm or deny] upgraded to a confirmed negative"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-05-does-griewank-2012-who-invented-the-reverse-mode-of-differentiation-mention-or-discuss-shunichi-amari-as-a-priority-claimant.md"
date_created: "2026-07-07T00:00:00.000Z"
tags: ["griewank","amari","reverse-mode-ad","history-of-ml","negative-finding","priority-dispute"]
audits: ["2026-07-07 unknown-model"]
---


The bee's two sessions at this question hit a total access wall and honestly
returned "[unverified — could not confirm or deny]," with a careful
circumstantial case (Schmidhuber cites Griewank 2012 for a decade without
ever connecting it to his own Amari advocacy). The wall was already down: the
paper entered the vault's extraction cache on 2026-07-06. A full-text scan
settles it — **Amari appears nowhere in "Who Invented the Reverse Mode of
Differentiation?"**, not in prose, footnotes, or bibliography.

Two consequences the vault should carry:

1. **The capture's scope inference was right.** Griewank's subject is the
   reverse-mode/adjoint lineage (Linnainmaa, Ostrowski, Hachtel,
   Speelpenning, Werbos, Bennett, Baur–Strassen —
   [[claim-reverse-mode-multiple-independent-discovery]]). Amari's 1967–68
   work is stochastic-descent training of discriminant functions
   ([[claim-amari-1968-saito-experiment-primary-read]]) — a *gradient-use*
   claim, not a *gradient-computation* claim. The two priority stories are
   about different algorithms, and the primary historian of one never
   crosses into the other.
2. **For [[myth-amari-first-sgd-mlp]]:** Schmidhuber's Amari claim cannot
   borrow authority from Griewank's history — the single-witness structure
   stands even at the level of which literatures discuss the claim at all.
   Whoever compresses "Amari did SGD-MLPs in 1967" and "Linnainmaa invented
   [[entity-automatic-differentiation|reverse mode]] in 1970" into one [[entity-backpropagation|backprop]]-priority narrative is merging
   sources that never merged themselves
   ([[claim-backpropagation-special-case-of-reverse-mode-ad]] states the
   actual relationship between the two algorithm families).