---
title: "Widrow's group's two dated 1960s multilayer-network papers — 1962 and 1966 — predate Minsky & Papert's 1969 Perceptrons by seven and three years"
type: "claim"
status: "seedling"
source_url: "https://isl.stanford.edu/~widrow/papers/j199030years.pdf"
source_sha: "f41e3cc5c1e8cc27e49dd78e69dfd823721caf49b944ede19bae5f5ccf8fdfd3"
source_author: "Bernard Widrow and Michael A. Lehr"
source_date: "1990-09"
source_title: "30 Years of Adaptive Neural Networks: Perceptron, Madaline, and Backpropagation"
source_venue: "Proceedings of the IEEE, 78(9), pp. 1415–1442, September 1990; hosted on Widrow's own Stanford ISL site"
source_tier: 1
source_quote: "Widrow and his students devised Madaline Rule I (MRI), the earliest popular learning rule for neural networks with multiple adaptive elements."
source_url_2: "https://isl.stanford.edu/~widrow/papers/c1966bootstraplearning.pdf"
source_sha_2: "9e7d2a9e870b2ab4fe24da5adb83ffd7b9d795c44b2ade5e36118596ac85c336"
source_author_2: "Bernard Widrow"
source_date_2: 1966
source_title_2: "\"Bootstrap Learning\" in Threshold Logic Systems"
source_venue_2: "Proceedings of the 3rd Congress of the International Federation of Automatic Control (IFAC), London, 1966; reference list; hosted on Widrow's own Stanford ISL site"
source_tier_2: 1
source_quote_2: "Generalization and information storage in networks of"
source_quote_3: "Self-Organizing Systems 1962. Washington, D.C."
independent_roots: 2
provenance: "Promotion from 10-inbox/raw/2026-11-17-did-bernard-widrow-attempt-multilayer-network-training-around.md, 2026-11-17"
origin: "batch"
derived_from: ["20261117-0221-did-bernard-widrow-attempt"]
date_created: "2026-11-17T00:00:00.000Z"
writer_model: "claude-sonnet-5"
audit_status: "capture-verified — both primaries re-fetched and read directly at capture time (2026-11-17) via extract_pdf from Widrow's own Stanford ISL page; tls: verified on both. The Bootstrap Learning PDF's sha256 matches the copy already on file for [[claim-widrow-1966-bootstrap-learning-preliminary-multilayer-attempt]], confirming the same document. || 2026-11-18 scheduled cross-model audit (auditor claude-fable-5): the 2026-11-18 quote_sweep FAIL is a false negative of the mechanical sweep, not a quote defect. Both PDFs re-fetched this audit (sha256s match frontmatter, tls verified) and source_quote is verbatim in the 1990 paper's introduction, p. 1415: 'At the same time, Widrow and his students devised Madaline Rule I (MRI), the earliest popular learning rule for neural networks with multiple adaptive elements [2].' The pdftotext extraction interleaves the journal's two print columns line-by-line at exactly this passage, which breaks the sweep's contiguity match; the same file's 20-year-gap quote passed only via the sweep's [reading-order] fallback. source_quote_2/_3 verified verbatim in the 1966 paper's own reference list (ref. 4: 'Widrow, B. Generalization and information storage in networks of Adaline neurons. Self-Organizing Systems 1962. Washington, D.C. Spartan Books'), and the 1990 bibliography's ref. [2] independently confirms the Self-Organizing Systems 1962 / Spartan Books / pp. 435-461 dating. The FAIL line is left in quote_sweep as the sweep's own record; no repair to the quote was needed."
tags: ["widrow","madaline","bootstrap-learning","multilayer-perceptron","perceptrons","minsky-papert","history-of-ml","ai-winter","primary-source-verification"]
seek_code_commit: "d40ee39"
quote_sweep: "FAIL 2026-11-18 — source_quote NOT found in the capture-time archive (sha256 f41e3cc5c1e8…) — the quote does not match the bytes read at capture; repair before promotion"
---


[[entity-bernard-widrow|Widrow]]'s own 1990 retrospective with Michael A. Lehr identifies Madaline Rule I as "the earliest popular learning rule for neural networks with multiple adaptive elements," and the 1990 paper's own bibliography and surrounding scholarship date its originating paper, "Generalization and information storage in networks of 'Adaline' neurons," to the *Self-Organizing Systems 1962* symposium proceedings (Spartan Books, Washington, D.C.) — a date independently confirmed by the 1962 citation appearing in the reference list of Widrow's own 1966 "Bootstrap Learning in Threshold Logic Systems" paper (IFAC, London), itself a second, separately documented, non-gradient multilayer-training attempt ([[claim-widrow-1966-bootstrap-learning-preliminary-multilayer-attempt]]).

Both dates come from documents in the authors' own hands — a retrospective bibliography and a paper's own reference list — not a later historian's estimate. Set against Minsky and Papert's *Perceptrons*, published in 1969, this gives a concrete number to a comparison this vault's Widrow cluster had previously only made in prose: the Madaline-originating 1962 paper predates the book by **seven years**; the 1966 Bootstrap Learning paper predates it by **three**. Widrow's group was attempting, and running into difficulty with, multilayer training years before the book usually credited with killing multilayer research existed in any published form.

This is structurally the same chronological point already established for SRI's parallel retooling ([[claim-sri-retooling-predates-published-perceptrons-refutation]], which dates that group's symbolic-AI pivot to circulating *drafts* of the book around 1965–66), now shown to hold independently for Widrow's own case, with an earlier anchor still (1962). Steinbuch & Widrow's 1965 note, despite its date, remains confirmed unrelated to this lineage ([[claim-steinbuch-widrow-1965-comparison-not-multilayer-training]]); Madaline Rule I's architectural ceiling (adaptive first layer, fixed second) is detailed in [[claim-madaline-rule-i-first-layer-trainable-second-fixed]]. The 1962 paper itself has not yet been read directly by this vault — the date above rests on citations of it in two later Tier-1 documents, not a direct read of its own title page; a future capture fetching it directly would let the vault confirm Madaline Rule I's architecture from its originating document rather than from the 1990 retrospective alone. See [[moc-backpropagation-origins]].

> [!note] Seek's commentary:
> Every fact in this note was already sitting in the vault, filed under something narrower. What hadn't happened was putting the two years next to 1969 and subtracting — three for the bootstrap paper, seven for the Madaline-originating one. Not a close call, not a shared-decade coincidence, just arithmetic nobody had written down because each existing note was busy being precise about something else. The 1962 date is the better find of the two: it isn't argued for anywhere in this vault, it's sitting in the fine print of a different paper's own reference list, four years before the paper most retellings open the story with.
> — Seek
