Does LeCun's 1985 paper actually derive backpropagation via 'duality theory from control theory (Bryson and Ho, 1969)', as Werbos's 2004 patent characterizes it?
claim-werbos-2004-object-nets-patent-names-lecun-against-1994-book quotes Werbos's 2004 "Object Nets" patent (US 6,708,160 B1) characterizing LeCun's 1985 derivation of backpropagation as reached "from an application of duality theory from control theory (Bryson and Ho, 1969)," then contrasting it with Werbos's own, broader 1974 derivation. This characterization appears only in a patent document written by Werbos himself, implicitly arguing for the priority and generality of his own method — a self-interested witness describing a rival's work, not an independent account.
What's needed: locate and read LeCun's 1985 paper directly (commonly cited as LeCun, "Une procédure d'apprentissage pour réseau à seuil asymmetrique," Cognitiva 85, Paris), and check whether it frames its derivation in control-theory/duality-theory terms at all, and whether it cites Bryson and Ho (1969) or any comparable control-theory source.
Why it matters: this is the one concrete, nameable primary that could confirm or overturn a mechanism claim currently resting entirely on one interested party's patent prose. It also bears on the vault's running Bryson–Ho control-theory-transmission thread (see claim-kelley-bryson-optimal-control-precursor, entity-arthur-bryson): a positive answer would be a second, independent instance of the Kelley–Bryson lineage reaching the neural-network community, this time via LeCun rather than Werbos.
Candidate next moves: search for an English translation or a close secondary reading of the 1985 Cognitiva paper (it is in French and not widely digitized); check whether Schmidhuber's 2014 survey or other backprop-historiography pieces describe its derivation method in more detail; if no digitized copy surfaces, treat as a manual-consultation item, the same disposition as the still-unlocated 1994 Werbos book.
Progress — 2026-10-16: Still unresolved at the primary-text level; the 1985 Cognitiva paper itself remains unlocated and unread. But LeCun's own research page, read directly, weighs against Werbos's specific attribution: LeCun describes the 1985 method as propagating "virtual target values," not gradients, with no control-theory or duality-theory framing stated anywhere on the page (claim-lecun-1985-virtual-target-values-not-gradients), and attributes the rigorous, Lagrangian (duality-adjacent) derivation specifically to his 1988 paper, three years later (claim-lecun-1988-lagrangian-derivation-not-1985). The likeliest explanation on current evidence is that Werbos's patent compresses LeCun's 1985 and 1988 publications into a single "LeCun (1985)" citation, not that the 1985 paper itself uses duality-theory language — but this remains an inference from LeCun's retrospective account, not a reading of the 1985 text. Candidate next move, not yet tried: check LeCun's own publications page (yann.lecun.com/exdb/publis/) via the Wayback Machine for a scanned copy of the 1985 paper before concluding none exists online.