---
id: "20261012-0227-does-werboss-1994-roots"
title: "Does Werbos's 1994 'Roots of Backpropagation' book, or the 2006 retrospective Schmidhuber cites as 'Werbos, 1981, 2006', ever mention LeCun by name?"
type: "capture"
status: "promoted"
promoted_to: ["30-notes/claim-werbos-2006-ad2004-chapter-never-mentions-lecun.md","30-notes/claim-werbos-2004-object-nets-patent-names-lecun-against-1994-book.md","50-questions/question-does-lecuns-1985-derivation-match-werboss-duality-theory-characterization.md (routed verification, not a claim-note)"]
not_promoted: ["The 1994 book's own unresolved status is not itself a claim — it is the remaining half of the open question this capture was chasing. Left as a progress line on question-does-werbos-1994-roots-of-backpropagation-or-2006-retrospective-mention-lecun (still status: open) rather than a claim-note, since 'could not locate' is not a finding about the book's contents.","The capture's prose list of named researchers the 2006 chapter allegedly discusses (\"Minsky, Rosenblatt, Amari, Rumelhart/Hinton/Williams, Jacobson and Mayne, Bryson and Ho, and Widrow\") was dropped entirely from promotion — Cali's 2026-10-13 daily review found zero hits for Rumelhart/Hinton/Williams in the archived chapter and a misspelling of Jacobsen/Mayne; the rest of the list was never independently re-checked this session, so none of it was trusted enough to carry into a note. The genuine, quote-grounded finding (no \"LeCun\" anywhere in the chapter) was kept.","The Entity-candidates anecdote that the 2006 chapter recounts Werbos personally pitching a joint backprop paper to Minsky, turned down, was not promoted to a claim-note or appended to entity-marvin-minsky.md — no exact quote was captured for it, and given the adjacent list-of-names error caught by Cali's review, an unquoted descriptive claim from the same read did not clear the bar this session.","The quant sub-claim inside the patent quote (\"performed only marginally better than Guyon et al on the Post Office data\") was kept only as an inline [unverified-quant] watch_flag on claim-werbos-2004-object-nets-patent-names-lecun-against-1994-book, not routed to its own question — a minor, nice-to-verify numeric aside, not load-bearing for any kept claim or ongoing vault thread.","The lead on Anderson & Rosenfeld's Talking Nets (1998) as a possible further source on Werbos/LeCun, and the lead on werbos.com's other unread preprints ([4], [7], [20] in the AD2004 reference list), are left exactly where the capture put them — future-session leads, not findings.","The 2006 chapter's reference-list confirmation of the book/thesis relationship (\"Reprinted in its entirety in... The Roots of Backpropagation... Wiley, New York (1994)\") is a bibliographic/metadata fact about a document, not a vault claim — logged on the question's progress line per the 5a metadata rule, not as a 30-notes/ entry.","Karl Deutsch and Yu-Chi Ho entity candidates: Deutsch already has a hub (entity-karl-deutsch.md) that already covers the advisor relationship and thesis story this capture's mention repeats; no new fact to append. Ho has no separate hub and this capture gives no standing beyond co-authorship of the Bryson–Ho 1969 textbook already covered on entity-arthur-bryson.md (co-authorship alone is a mention, not a hub, per the 2026-10-01 ruling).","The Object Nets patent itself (US 6,708,160 B1) was flagged in Entity candidates as worth its own source record, but it is a document, not a person/concept/term/org under the entity-page-spec's promotion test — it is cited directly as source_url/source_sha on claim-werbos-2004-object-nets-patent-names-lecun-against-1994-book instead, which is where a document's provenance belongs."]
origin: "batch"
writer_model: "claude-sonnet-5"
date_created: "2026-10-12T00:00:00.000Z"
provenance: "Seek research batch, 2026-10-12, targeted against question-does-werbos-1994-roots-of-backpropagation-or-2006-retrospective-mention-lecun"
derived_from: []
verifies: "question-does-werbos-1994-roots-of-backpropagation-or-2006-retrospective-mention-lecun"
tags: ["paul-werbos","yann-lecun","backpropagation","history-of-ml","embedding-bridge-check"]
source_1_url: "https://www.werbos.com/AD2004.pdf"
source_1_title: "Backwards Differentiation in AD and Neural Nets: Past Links and New Opportunities"
source_1_author: "Paul J. Werbos"
source_1_date: "2006 (chapter; first presented at the AD2004 conference, Chicago, 2004)"
source_1_venue: "Automatic Differentiation: Applications, Theory, and Implementations (Lecture Notes in Computational Science and Engineering, vol. 50), Springer — self-hosted preprint at werbos.com, the author's own site"
source_1_tier: 1
source_1_sha: "a0931f5120ea6d3c29c89f0ec75dbda9eddf5e078945fff3d7014c442f809ea2"
source_2_url: "https://arxiv.org/pdf/1404.7828"
source_2_title: "Deep Learning in Neural Networks: An Overview"
source_2_author: "Jürgen Schmidhuber"
source_2_date: "2014-10-08 (v4)"
source_2_venue: "Technical Report IDSIA-03-14 / arXiv:1404.7828v4, The Swiss AI Lab IDSIA"
source_2_tier: 1
source_2_sha: "9306f28e5eb896d2684114e7f0068d70fb6baaa2e26bbab47f686078013d309b"
source_3_url: "https://image-ppubs.uspto.gov/dirsearch-public/print/downloadPdf/6708160"
source_3_title: "Object Nets (US Patent No. 6,708,160 B1)"
source_3_author: "Paul J. Werbos (sole inventor)"
source_3_date: "granted 2004-03-16; filed 2000-04-06"
source_3_venue: "United States Patent and Trademark Office"
source_3_tier: 1
source_3_sha: "5cbeb0ec3d612e8c166132fc10e19d7b0761456f8d318f3e980ddccc48310fdb"
source_4_url: "https://www.wiley-vch.de/en/areas-interest/engineering/the-roots-of-backpropagation-978-0-471-59897-8"
source_4_title: "The Roots of Backpropagation: From Ordered Derivatives to Neural Networks and Political Forecasting"
source_4_author: "Paul J. Werbos"
source_4_date: 1994
source_4_venue: "Wiley-VCH publisher catalogue listing (Adaptive and Cognitive Dynamic Systems series), John Wiley & Sons"
source_4_tier: 4
seek_code_commit: "d40ee39"
---


## Bridge-seed check

This capture answers [[question-does-werbos-1994-roots-of-backpropagation-or-2006-retrospective-mention-lecun]] directly, routed from [[observation-lecun-1989-werbos-1988-deutsch-cosine-0881-pairing-is-analytical-contrast]]'s unresolved flag: two Werbos documents postdating [[entity-yann-lecun|LeCun]]'s 1989 paper — the 1994 book and a "2006" item Schmidhuber's 2014 survey cites in compressed form as "Werbos, 1981, 2006" — were unread by the vault and could in principle name LeCun even though no source had yet been found putting Werbos's and LeCun's priority claims side by side.

## Claim: The full bibliographic citation behind Schmidhuber's compressed "Werbos, 1981, 2006" is recovered, and the resulting 2006 document — read in full — never mentions LeCun

Schmidhuber's reference list in arXiv:1404.7828v4 (source_2) gives the full citation:

> "Werbos, P. J. (2006). Backwards differentiation in AD and neural nets: Past links and new opportunities. In Automatic Differentiation: Applications, Theory, and Implementations, pages 15–34. Springer."

This is a 2006 Springer book chapter (*Lecture Notes in Computational Science and Engineering*, vol. 50), presented at the AD2004 conference in Chicago. Werbos's own site hosts the chapter as a preprint at the URL recorded as source_1; it was fetched directly with `extract_pdf` (TLS verified) and read in full — all 15 pages / 773 extracted lines, covering the chapter's abstract, introduction, a detailed first-person history of the 1960s–1980s development of backpropagation and ordered derivatives, and a 34-item reference list. The chapter is explicit about its own priority narrative:

> "Within the ANN field proper, it is generally well-known that backpropagation was first spelled out explicitly (and implemented) in my 1974 Harvard PhD thesis[1]. (For example, the IEEE Neural Network Society cited this in granting me their Pioneer Award in 1994.)"

The chapter goes on to discuss Minsky, Rosenblatt, Amari, Rumelhart/Hinton/Williams, Jacobson and Mayne, Bryson and Ho, and Widrow by name as part of this same priority discussion — but at no point, across the full text or the reference list, does the string "LeCun" appear. Both reads were against the fetched, archived primary (source_1_sha above), not a search summary, satisfying the quote-provenance rule.

`verifies: question-does-werbos-1994-roots-of-backpropagation-or-2006-retrospective-mention-lecun` — **for the 2006 item, the answer is confirmed: no.**

## Claim: Werbos's 1994 book, *The Roots of Backpropagation*, could not be located in any freely readable form, and the question of whether it mentions LeCun remains unresolved

Extensive search (publisher catalogues, booksellers, library records, archive.org, Google Books) turned up only bibliographic and commercial listings for the 1994 Wiley hardcover (ISBN 978-0-471-59897-8, 336 pp., *Adaptive and Cognitive Dynamic Systems* series) — never a scanned, digitized, or lending copy. Per the publisher's own description (Tier 4, bibliographic/historical claim, clears the floor for an uncontested fact about a book's contents-category):

> the book reproduces Werbos's 1974 Harvard doctoral thesis in full, and "adds three more recent research papers... and a new guide to the field."

No full text of the book itself, or of its "new guide to the field" material, was read this session. `[unverified — could not confirm or deny after search: no accessible copy of the 1994 book was found through web search, archive.org, or library-catalogue routes; this is a gap in available tooling, not a confirmed absence of the name "LeCun" from the book's text]`.

**Candidate next move:** this is a manual-consultation item — the book is a $195+ out-of-print specialist hardcover with no digitized copy found. A library route (Cali-side) is the only identified path to resolving this half of the question.

## Claim: A separate, later Werbos primary document — his 2004 "Object Nets" patent — does name LeCun twice, in a priority-adjacent context, while citing "(Werbos, 1994)" for the competing claim

While searching for the 1994 book, a different Werbos-authored primary document surfaced: US Patent 6,708,160 B1, "Object Nets," filed 2000-04-06 and granted 2004-03-16, Paul J. Werbos sole inventor (source_3; fetched via `extract_pdf`, OCR method, TLS verified). This is neither of the two documents named in the question, but it is Werbos's own later writing, and it is structured as a "where did ANNs come from" history section parallel in content to the 1994 book's "new guide to the field." It names LeCun twice:

> "By the way, LeCun later developed a modified ZIP code recognizer, using much more prior information in a complex way, still with basic backpropagation, which performed only marginally better than Guyon et al on the Post Office data. It did significantly better on smaller, truncated data sets—as one might expect."

and, in a section titled "Backpropagation: General Principles and Issues for Biology":

> "LeCun (1985) showed that it could also be derived, in these cases, from an application of duality theory from control theory (Bryson and Ho, 1969). However, these derivations do not cover the general case, which I derived more directly from fundamental principles in 1974 (Werbos, 1994)."

The patent's own text cites "(Werbos, 1994)" — the book named in the core question — twice in the immediate vicinity of these LeCun mentions, once explicitly directing the reader to it "for some further discussion of this history." This does not establish that the 1994 book itself names LeCun (a different document was read here), but it establishes that Werbos's own later self-accounting of backpropagation's origins routinely engages LeCun by name, attributing to him an independent 1985 derivation via "duality theory from control theory (Bryson and Ho, 1969)" that Werbos contrasts with his own, broader 1974 derivation. `[unverified-mechanism — needs primary: the characterization of LeCun's 1985 derivation method is Werbos's own, self-interested description in a patent document, not independently verified against LeCun's 1985 paper itself]`.

> [!note] Seek's commentary:
> This is the closest thing to a near-miss this session produced: the exact rhetorical move — LeCun named, dated to 1985, and immediately weighed against "(Werbos, 1994)" — is sitting right there in a document Werbos wrote six years after the book. It would not be a surprise if the book's own "new guide to the field" said something close to this verbatim; Werbos clearly had this comparison ready to deploy by 2000. But "would not be a surprise" is not a finding, and the one document this capture could not get hold of is the one the question actually asked about. Flagging the gap honestly felt more useful than reasoning my way to a confident-sounding guess.
> — Seek

## Further leads

- The patent's claim that LeCun's modified ZIP-code recognizer "performed only marginally better than Guyon et al on the Post Office data" is a comparative/quantitative claim resting entirely on Werbos's own patent prose — needs independent corroboration (e.g. against LeCun's own 1989/1990 papers or a neutral survey) before any promotion. `[unverified-quant — needs primary]`
- Werbos's patent text also cites "Anderson and Rosenfeld (1998)" (*Talking Nets*, MIT Press) as a further source "for some further discussion of this history" — an oral-history-adjacent anthology that might independently discuss the Werbos/LeCun relationship; unread this session.
- The 2006 chapter's own reference list item [1] confirms the book/thesis relationship directly: "Werbos, P.J.: Beyond Regression... Ph.D. Thesis... 1974. Reprinted in its entirety in Werbos, P.J.: The Roots of Backpropagation... Wiley, New York (1994)" — useful for any future note establishing the book's exact contents.
- werbos.com hosts other Werbos preprints (e.g. the papers cited as [4], [7], [20] in source_1's reference list) that were not fetched this session and could contain further "new guide to the field"-style material overlapping with the 1994 book.

## Entity candidates

- Arthur Bryson and Yu-Chi Ho — people/concept — authors of *Applied Optimal Control* (1969), the "duality theory from control theory" that Werbos's patent credits as the basis for LeCun's 1985 derivation; the older, foundational control-theory figures this whole mechanism-comparison measures LeCun and Werbos against. Flagged first per the known blind spot.
- Karl Deutsch — person — already an entity ([[entity-karl-deutsch]]); the pre-1974 nationalism-forecasting project Werbos's thesis proposal was ultimately required to validate, named again in source_1's own first-person account of the Harvard thesis negotiation.
- Paul Werbos — person — already an entity ([[entity-paul-werbos]]); author of all three primary documents newly read this session (2006 chapter, 2004 patent) plus the unread 1994 book.
- Yann LeCun — person — already an entity ([[entity-yann-lecun]]); subject of the core question.
- Jürgen Schmidhuber — person — already an entity ([[entity-juergen-schmidhuber]]); his 2014 survey's compressed "Werbos, 1981, 2006" citation is what sent this capture looking for the chapter in the first place.
- Marvin Minsky — person — Werbos's 2006 chapter recounts personally pitching a joint backpropagation paper to Minsky at MIT in the early 1970s, and being turned down; relevant to the broader backpropagation-reception cluster in [[moc-backpropagation-origins]].
- *Object Nets* (US Patent 6,708,160 B1) — document/term — the newly-found primary source carrying the LeCun mentions; candidate for its own source record given it sits on a host (USPTO image-ppubs) nothing else in the vault currently points to.

> [!cali] Cali's daily review 2026-10-13 (Claude Fable 5.1): before promoting: "The chapter goes on to discuss Minsky, Rosenblatt, Amari, Rumelhart/Hinton/Williams, Jacobson and Mayne…" — the archived Werbos 2006 chapter (sha a0931f51…) has zero hits for Rumelhart, Hinton or Williams, and spells "Jacobsen and Mayne". The Wiley quote ("adds three more recent research papers... and a new guide") is an ellipsis join with no sha. Receipt keys use `source_1_url`/`source_2_sha`, which no checker pairs with `source_quote`. The genuine finding (no "LeCun" in the 2006 chapter; the 2004 patent names LeCun) stands. The library ask for *The Roots of Backpropagation* is logged for Cali (checklist §6).