“But he neither invented them nor was he the first to apply the back-propagation algorithm to learning their weights (though many people mistakenly believe this).”
Our reading, the sentence that was checked:
LeCun neither invented convolutional neural networks nor was the first to train their weights with back-propagation.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, The False Glorification of Yann LeCun, paragraph 8.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-the-false-glorification-of-yann-lecun-f48859-a1
We recorded 7 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2025-11-18. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
From the speaker
- neither invented them
- nor was he the first to apply the back-propagation algorithm to learning their weights
- convolutional neural networks
Our interpretation
- CNN refers to the broad architecture family, including shared-weight temporal networks as well as image networks.
- Foundational architecture, backpropagation training, and later practical improvements are distinct contributions.
- The historical comparison covers the early development described in the context, through 1989.
Counts as supported if: Dated records establish a convolutional architecture before LeCun's work and backpropagation training of a qualifying convolutional network before his first such work.
Counts as contradicted if: The alleged predecessors lack the relevant architecture or training method, and reliable chronology establishes LeCun's priority for either asserted first.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The broad architecture reading is supported. Fukushima's neocognitron predates LeCun's work. Waibel's October 1987 report describes a shared-weight temporal network trained with backpropagation, while LeCun dates the start of his shared-weight experiments to 1988. This establishes earlier work for both parts without establishing Zhang as the first researcher to combine CNNs and backpropagation.
Limitations
- This result includes temporal convolutional networks; a claim restricted to two-dimensional image CNNs would require a separate priority comparison.
- LeCun's experimental starting date comes from a retrospective interview available through a reproduction.
- The result does not deny LeCun's original improvements or practical contributions.
- The quotation from the 1987 scan preserves its extraction artifacts.
Evidence (3 items)
Supports the claim: IEICE NOLTA; Kunihiko Fukushima,
hierarchical multi-layered network.
Why it matters here: Fukushima identifies the neocognitron as a deep CNN introduced in 1979, with references to the 1979 Japanese and 1980 English publications. This supports the foundational-architecture part of the claim.
- Publisher
- IEICE NOLTA; Kunihiko Fukushima
- Kind of source
- Peer-reviewed paper
- Describes events of
- Public from
Supports the claim: ATR Interpreting Telephony Research Laboratories; KIT-hosted copy,
Back-propagation Learning Procedure{20:5].
Why it matters here: The report describes backpropagation and constraining corresponding connections across time shifts to share weights. It supplies a temporal convolutional network trained by backpropagation before the 1988 start described by LeCun.
- Publisher
- ATR Interpreting Telephony Research Laboratories; KIT-hosted copy
- Kind of source
- Preprint, not peer reviewed
- Describes events of
- Public from
Supports the claim: Kaggle interview, reproduced by GarfieldEr007 on CNBlogs,
in Toronto in 1988.
Why it matters here: LeCun's retrospective account places the beginning of his shared-weight experiments after Waibel's October 1987 report. The available_at date is the dated reproduction, not the original interview date.
- Publisher
- Kaggle interview, reproduced by GarfieldEr007 on CNBlogs
- Kind of source
- News report
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, The False Glorification of Yann LeCun
But he neither invented them nor was he the first to apply the back-propagation algorithm to learning their weights (though many people mistakenly believe this).
- Where
- paragraph 8
- Published
- 2025-11-18
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-the-false-glorification-of-yann-lecun-f48859-a1: Current. Outcome: Supported. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-the-false-glorification-of-yann-lecun-f48859-a1.