---
title: "Do Steinbuch & Widrow 1965 and Widrow's 1966 'Bootstrap Learning' actually document the abandoned multilayer-training attempt — and what does Widrow's oral history say about the 1985 return?"
type: "question"
status: "open"
date_raised: "2026-07-09T00:00:00.000Z"
tags: ["widrow","madaline","multilayer-perceptron","backpropagation","history-of-ml","primary-source-verification"]
---


Raised while promoting [[claim-widrow-abandoned-multilayer-training-until-1985-backprop]]
and [[claim-madaline-rule-i-first-layer-trainable-second-fixed]]. The mechanism
and the Madaline Rule I architecture are now Tier-1 confirmed from Widrow & Lehr
1990 (read directly once the poppler/`pdfinfo` dependency was repaired,
2026-07-09). Two verifications remain:

1. **The 1965–66 date-coincidence.** Two real Widrow papers fall in the window
   named by the original research question but their link to "the" abandoned
   attempt is unverified (the capture flagged this as a date-coincidence trap):
   - Steinbuch & Widrow, "A Critical Comparison of Two Kinds of Adaptive
     Classification Networks," IEEE Trans. Electronic Computers EC-14(5):737–740,
     1965 — https://isl.stanford.edu/~widrow/papers/j1965acritical.pdf
   - Widrow, "Bootstrap Learning in Threshold Logic Systems," 3rd IFAC Congress,
     London, 1966 — https://isl.stanford.edu/~widrow/papers/c1966bootstraplearning.pdf
   Read each directly (the extract path now works — Widrow's other scanned PDFs
   extracted cleanly via pdftotext) and determine whether either paper is about a
   multilayer-training attempt or an adjacent topic.

2. **The 1985 Snowbird return.** The specific "1985 Snowbird, Utah" reconnection
   to backpropagation currently rests on Yuxi Liu (T2) and Wikipedia (T4). The
   primary is Widrow's oral history (https://ethw.org/Oral-History:Bernard_Widrow,
   403-blocked at capture time) or his 2023 "Ancient History" chapter. Confirm the
   date/venue and the "sharp quantizers / sigmoids / no one knew anything about it"
   quote against a directly-read primary.

Low-to-medium priority — the load-bearing claims are already primary-verified; this
tightens the biographical precision and closes a flagged date-coincidence.