Widrow & Lehr's 1990 retrospective states the group's return to neural-network research followed a self-reported 20-year gap, corroborating but not confirming the mid-1980s return window
In its own introduction, Widrow & Lehr's 1990 retrospective states: "After 20 years of research in adaptive signal processing, the work in Widrow's laboratory has once again returned to neural networks." Counted forward from the mid/late-1960s point at which the group's attempts to train multiple adaptive layers were unsuccessful and work "switched to adaptive filtering and adaptive signal processing" — already Tier-1-established in claim-widrow-abandoned-multilayer-training-until-1985-backprop from the same document — a self-reported 20-year gap lands in the mid-to-late 1980s. This is consistent with, though does not use, the same words as the "1985 Snowbird, Utah" detail that question-verify-widrow-1965-66-papers-multilayer-attempt still flags as resting on Wikipedia (Tier 4) alone. It is also consistent with the same paper's later statement that "In 1987, Widrow, Winter, and Baxter looked back at the original Madaline I algorithm" to build Madaline Rule II.
This sentence names neither "1985" nor "Snowbird" and so does not close the open question about that specific date and venue — it is recorded here as Tier-1 corroboration, in the authors' own words, of the decade of the return, independent of the still-unread Talking Nets (2000) and Magoun's 2014 Proceedings of the IEEE piece that question's log names as the most likely primary home of a dated account. See claim-madaline-rule-i-first-layer-trainable-second-fixed and moc-backpropagation-origins.
Source
“After 20 years of research in adaptive signal processing, the work in Widrow's laboratory has once again returned to neural networks.”
claude-sonnet-5 · Promotion from 10-inbox/raw/2026-11-17-did-bernard-widrow-attempt-multilayer-network-training-around.md, 2026-11-17 · raw markdown