---
title: "Bernard Widrow"
type: "entity"
entity_kind: "person"
status: "hub"
canonical_name: "Bernard Widrow"
aliases: ["Widrow"]
first_seen: "2026-07-09T00:00:00.000Z"
writer_model: "claude-sonnet-5"
connects_to: ["LMS algorithm","Adaline / Madaline","Ted Hoff","Robert W. Lucky","adaptive signal processing"]
seek_code_commit: "a619c8a"
---


Stanford engineer, co-inventor with Ted Hoff of the LMS algorithm and the
Adaline (early 1960s) — the origin point of the vault's adaptive-filter and
credit-assignment lineages. Unable to extend gradient training past a single
layer with the era's hard-limiting quantizers, his group left neural-network
research for adaptive signal processing (antennas, echo cancelling, seismic
processing) until backpropagation reached him in the mid-1980s. Load-bearing
across two of the vault's clusters: the backprop-origins/multilayer-training
thread, and — via his own oral history and retrospective papers — the
Lucky/Bell-Labs multiple-discovery pair in adaptive equalization. Recurs
across enough claim-notes (six-plus, since 2026-07-09) that this hub was
overdue — part of the general pre-2026-07-14 entity-page backlog tracked in
`00-meta/seek-flags.md`.

## References
- [[claim-widrow-abandoned-multilayer-training-until-1985-backprop]] · [[claim-madaline-rule-i-first-layer-trainable-second-fixed]] · [[claim-ted-hoff-widrow-phd-student-architected-intel-4004]] · [[claim-amari-stochastic-descent-and-widrow-hoff-lms-are-sgd-siblings]] · [[claim-lucky-1965-adaptive-equalizer-steepest-descent-transversal-filter]]
- [[claim-widrow-frames-lucky-equalizer-as-independent-invention-lms-later-graft]] · [[claim-lucky-widrow-independent-origination-favored-lucky-side-unverified]] · [[observation-persistent-homology-gradient-free-bridge-widrow-byzantine]]
- [[moc-backpropagation-origins]] · [[question-lucky-widrow-adaptive-filter-mutual-awareness]] · Captures: 2026-07-08-did-widrow-attempt-and-abandon-multilayer-network-training-around-1965-1966, 2026-07-19-were-luckys-bell-labs-and-widrows-stanford-adaptive
- 2026-08-15: his own 1960 (with Hoff) and 1990 (with Lehr) papers, read directly, establish the LMS/Adaline mechanism as stochastic gradient descent in Widrow's own words — the earliest primary use of "stochastic" for the error surface being searched, and the 1990 proof that the instantaneous gradient is an unbiased estimate of the true gradient. ([[claim-widrow-hoff-1960-original-paper-describes-lms-as-stochastic-steepest-descent]], [[claim-widrow-lehr-1990-lms-instantaneous-gradient-unbiased-estimate]])
- 2026-09-05: three more primaries read directly, sorting a tangled biographical story. His 1965 paper with Karl Steinbuch turns out to be an unrelated capacity comparison, not a multilayer-training study; his 1966 "Bootstrap Learning" paper is a genuine, separate, non-gradient attempt at multilayer training — a second 1960s attempt alongside Madaline Rule I; and his full 1997 oral-history transcript, read end to end, does not contain the "1985 Snowbird" return story commonly attributed to him — that detail now traces to Wikipedia alone among sources checked. ([[claim-steinbuch-widrow-1965-comparison-not-multilayer-training]], [[claim-widrow-1966-bootstrap-learning-preliminary-multilayer-attempt]], [[claim-widrow-abandoned-multilayer-training-until-1985-backprop]])