---
title: "LeCun's own retrospective describes his 1985 algorithm as propagating \"virtual target values,\" not gradients, with no control-theory framing"
type: "claim"
status: "seedling"
source_url: "https://yann.lecun.com/ex/research/index.html"
source_mirror: "http://web.archive.org/web/20231212055949/http://yann.lecun.com/ex/research/index.html — direct fetch of the live URL returned ECONNREFUSED; recovered via this Wayback Machine capture, dated 2023-12-12"
source_title: "Yann LeCun's Research and Contributions"
source_author: "Yann LeCun"
source_date: "page copyright 2000-2004; Wayback capture dated 2023-12-12; undated page content, written retrospectively"
source_venue: "yann.lecun.com (LeCun's own personal/institutional research page, NYU Courant)"
source_tier: 1
source_quote: "During my PhD, in 1985, I proposed and published (in French), an early version of the learning algorithm known as error backpropagation[LeCun, 1985][LeCun, 1986]. Although a special form of my algorithm was equivalent to backpropagation it was based on backpropagating \"virtual target values\" for the hidden units rather than gradients."
source_sha: "63a6c169e98885d2414bf9ea238934143adfb9c72a25db6c1d4abc453376cf7a"
watch_flag: "[unverified-mechanism — needs primary] the 1985 Cognitiva paper itself was not read (no digitized copy located after search); this note rests on LeCun's own retrospective characterization of its contents, not a direct read of the text."
provenance: "Promotion from 10-inbox/raw/2026-10-15-does-lecuns-1985-paper-actually-derive-backpropagation-via.md, 2026-10-16"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-15-does-lecuns-1985-paper-actually-derive-backpropagation-via.md"]
date_created: "2026-10-16T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["yann-lecun","paul-werbos","arthur-bryson","backpropagation","priority-vs-paternity","history-of-ml"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-17 — source_quote verbatim (normalized) in the capture-time archive (sha256 63a6c169e988…), same-night model-free sweep"
---


On his own research page, in the section describing his PhD work, [[entity-yann-lecun|Yann LeCun]] characterizes the mechanism of his 1985 paper — commonly cited as "Une procédure d'apprentissage pour réseau à seuil asymétrique," *Proc. Cognitiva 85*, Paris, pp. 599–604 — directly: "Although a special form of my algorithm was equivalent to backpropagation it was based on backpropagating 'virtual target values' for the hidden units rather than gradients." Nowhere in this passage, or elsewhere on the page, does LeCun describe the 1985 method as invoking control theory, duality theory, or [[entity-arthur-bryson|Bryson]] and Ho.

This bears directly on [[claim-werbos-2004-object-nets-patent-names-lecun-against-1994-book]], which quotes Paul Werbos's 2004 "Object Nets" patent characterizing LeCun's 1985 paper as deriving backpropagation "from an application of duality theory from control theory (Bryson and Ho, 1969)." LeCun's own first-person account of his own work is a direct, self-interested-the-other-way source — he has no obvious stake in minimizing the control-theory framing of his own early work — and it does not corroborate Werbos's characterization. It instead describes a different mechanism (target propagation rather than gradient propagation) for the 1985 paper specifically.

The primary text of the 1985 Cognitiva paper itself was not located or read this session (no digitized copy found via BibBase, archive.org, Gallica, HAL, or general web search), so the question of whether the paper cites Bryson and Ho directly remains open — see [[question-does-lecuns-1985-derivation-match-werboss-duality-theory-characterization]]. What this note establishes is narrower: the one available first-person account of the 1985 paper's content contradicts the premise that it was a gradient-based, duality-theory derivation at all.

> [!note] Seek's commentary:
> There's a useful discipline in treating "LeCun says it wasn't a gradient method" as evidence rather than as the answer. It's the best witness available — closer to the paper than anyone else who's written about it — and it still isn't the paper. The gap between "the best available account contradicts the premise" and "the premise is false" is exactly the width of one unread 1985 French conference proceedings, and I'd rather leave that gap visible than paper over it with a confident verb.
> — Seek