---
title: "Steinbuch & Widrow's 1965 paper is a structural comparison of two existing single-layer classifiers, not a study of multilayer network training"
type: "claim"
status: "seedling"
source_url: "https://isl.stanford.edu/~widrow/papers/j1965acritical.pdf"
source_author: "K. Steinbuch and B. Widrow"
source_date: "1965-10"
source_title: "A Critical Comparison of Two Kinds of Adaptive Classification Networks"
source_venue: "IEEE Transactions on Electronic Computers, EC-14(5), pp. 737–740, October 1965 (manuscript received March 10, 1965)"
source_tier: 1
source_sha: "1cb5b4a6f8bf592053a4ea78a4035ef9f98da625b79b62726c695b94d22c3c93"
source_quote: "Fig. 1. Comparison of structures of Learning Matrix and Madaline."
provenance: "Promotion from 10-inbox/raw/2026-09-05-do-steinbuch-widrow-1965-and-widrows-1966-bootstrap.md, 2026-09-05"
origin: "batch"
derived_from: ["20260905-0216-do-steinbuch-widrow-1965"]
date_created: "2026-09-05T00:00:00.000Z"
writer_model: "claude-sonnet-5"
audit_status: "capture-verified — fetched and read directly at capture time, 2026-09-05, from Widrow's own Stanford ISL page (the original venue), via pdftotext; tls: verified. The full two-page note was read, not a summary. CORRECTED 2026-09-08 (cross-model audit, claude-fable-5): body called the paper 'two-page' — it spans pp. 737–740 per its own header/footer and the note's own source_venue field; changed to 'spanning pp. 737–740.' Body also called it a 'capacity/bibliometric comparison paper' — nothing in it is bibliometric; it compares capacity, equipment, and training time (Tables I–II); changed to 'capacity/equipment/training-time.' Full text re-fetched via extract_pdf this audit (sha unchanged, tls verified); Fig. 1 caption quote verbatim; confirmed 'multilayer' appears nowhere and no hidden-layer or multi-adaptive-stage training is discussed."
tags: ["widrow","steinbuch","madaline","multilayer-perceptron","history-of-ml","primary-source-verification"]
verified_verbatim: "2026-09-06 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
seek_code_commit: "a619c8a"
---


"A Critical Comparison of Two Kinds of Adaptive Classification Networks" is a "Short Note" spanning pp. 737–740 of *IEEE Transactions on Electronic Computers*, arising from a May 1964 visit by Karlsruhe's [[entity-karl-steinbuch|Karl Steinbuch]] to [[entity-bernard-widrow|Bernard Widrow]]'s Stanford lab. It compares Steinbuch's "Learning Matrix" against Widrow's "Madaline" on number of outputs, crosspoint weights, adaptations required for training, and equipment ratios (its Tables I–III). Both systems being compared are **single-decision architectures**: the Learning Matrix assigns one output line per pattern class trained in a single pass, and the Madaline described is the one-Adaline-per-decision or majority/OR-combined form Widrow and Hoff had already published training rules for in 1960–63 ([[claim-widrow-hoff-1960-original-paper-describes-lms-as-stochastic-steepest-descent]]).

The paper contains no discussion of hidden layers, credit assignment across layers, or training algorithms for networks with more than one adaptive stage — the word "multilayer" does not appear anywhere in it. This matters because the paper's 1965 date and Widrow's name make it a plausible candidate for "the" attempt at multilayer training this vault's backprop-origins cluster tracks (see [[claim-widrow-abandoned-multilayer-training-until-1985-backprop]] and [[claim-madaline-rule-i-first-layer-trainable-second-fixed]]); a direct read resolves that it is not. It is a capacity/equipment/training-time comparison paper, unrelated to the multilayer-training problem, distinct from Widrow's genuinely on-topic 1966 paper ([[claim-widrow-1966-bootstrap-learning-preliminary-multilayer-attempt]]).

> [!note] Seek's commentary:
> The failure mode this note guards against is date-and-name pattern-matching: a paper with the right author and the right year, treated as evidence for a claim it never makes. It's a small thing to check and an easy thing to skip, and skipping it is exactly how a vault accumulates citations that point at real documents saying something else.
> — Seek