---
title: "Baydin et al. 2018 hedge Linnainmaa's priority within a multi-figure lineage and defer the reverse-mode origin question to Griewank (2012)"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/pdf/1502.05767"
source_sha: "6f72720b86da67c0471efb879677e3909cdf32ff4d73c006663b8c441a4cf552"
source_title: "Automatic Differentiation in Machine Learning: a Survey"
source_author: "Baydin, Pearlmutter, Radul, Siskind"
source_date: "2018-02-05T00:00:00.000Z"
source_venue: "arXiv preprint v4 (also published as Journal of Machine Learning Research 18(153):1-43, 2018)"
source_tier: 1
source_quote: "This method was understood in the control theory community (Bryson and Denham, 1962; Bryson and Ho, 1969) and cast in more formal terms with discrete-time variables topologically sorted in terms of dependency by Werbos (1974). Prior to Werbos, the work by Linnainmaa (1970, 1976) is often cited as the first published description of the reverse mode."
source_quote_2: "For a thorough review of the development of AD, we advise readers to refer to Rall (2006). Interested readers are highly recommended to read Griewank (2012) for an investigation of the origins of the reverse mode and Schmidhuber (2015) for the same for backpropagation."
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-11-21-what-does-baydin-et-al-2018-arxiv-150205767.md, 2026-11-21"
origin: "batch"
derived_from: ["10-inbox/raw/2026-11-21-what-does-baydin-et-al-2018-arxiv-150205767.md"]
date_created: "2026-11-21T00:00:00.000Z"
tags: ["automatic-differentiation","linnainmaa","baydin-2018","backpropagation","history-of-ml","reverse-mode-ad","werbos","griewank","historiography-of-ml"]
audit_status: "capture-verified — full direct extract_pdf read of the 43-page arXiv v4 PDF (sha256 above, TLS verified), §3.3 'Origins of AD and Backpropagation' (p.13-14) read in full for both quotes. Queen's independent re-fetch not performed — no network access in this headless promotion, by design."
seek_code_commit: "d40ee39"
---


[[entity-automatic-differentiation|Automatic differentiation]]'s most-cited survey — Baydin, Pearlmutter, Radul & Siskind's "Automatic Differentiation in Machine Learning: a Survey" (arXiv:1502.05767, also JMLR 18(153), 2018) — names [[entity-seppo-linnainmaa|Seppo Linnainmaa]] in its §3.3, "Origins of AD and Backpropagation," but does not treat him as the sole or settled origin of reverse-mode AD.

The section narrates a sequence: the Pontryagin maximum principle (Rozonoer 1959; Boltyanskii et al. 1960, the lineage headed by [[entity-lev-pontryagin|Pontryagin]]) as "the essence of the reverse mode" in continuous time; the control-theory community (Bryson & Denham 1962; Bryson & Ho 1969); then [[entity-paul-werbos|Werbos]] (1974), credited with casting the method "in more formal terms with discrete-time variables topologically sorted in terms of dependency" — narrated immediately *before* Linnainmaa even though Linnainmaa's 1970 thesis predates Werbos's 1974 dissertation by four years. Only then does the hedge appear:

> "Prior to Werbos, the work by Linnainmaa (1970, 1976) is often cited as the first published description of the reverse mode."

"Is often cited as" attributes the claim to a citation convention the field has adopted, not to the survey's own adjudicated verdict — a distinction already visible in how [[entity-andreas-griewank|Griewank]]'s own 2012 and 2014 papers hedge the same point rather than asserting it ([[claim-griewank-2012-full-text-contains-no-priority-but-no-paternity-phrase]], [[claim-griewank-2014-names-no-werbos-or-rumelhart-defers-history-to-2012-essay]]). The survey confirms this reading by explicitly declining to adjudicate the question itself, closing its historical section with:

> "Interested readers are highly recommended to read Griewank (2012) for an investigation of the origins of the reverse mode and [[entity-juergen-schmidhuber|Schmidhuber]] (2015) for the same for backpropagation."

So the field's own standard-reference survey treats "who originated reverse-mode AD" as a question for dedicated historical papers to answer, not as settled fact it will assert in its own voice — consistent with [[claim-reverse-mode-multiple-independent-discovery]] and the [[claim-linnainmaa-priority-not-paternity|priority-without-paternity]] framing the vault already carries for this cluster.

> [!note] Seek's commentary:
> What strikes me is the order of operations: the survey puts Werbos's later dissertation ahead of Linnainmaa's earlier thesis in the sentence, then softens the one clause that does credit Linnainmaa with "is often cited as" — as if sequence and hedge strength are both doing quiet editorial work the prose never has to own up to. A survey can look perfectly neutral while still nudging a reader's sense of who came first.

## Vault connections
Sits beside [[claim-ml-and-ad-communities-mutually-unaware]] (same survey, same §3.3 sentence, a different claim about community structure rather than origin-adjudication) and its companion note [[claim-baydin-et-al-2018-splits-credit-linnainmaa-description-speelpenning-implementation]] (same section, the Linnainmaa/Speelpenning credit-split half of this capture's finding).
