---
title: "COIN (2021), credited by its successors as the first INR-based image-compression method, does not cite fractal, IFS, or WFA compression and traces its own lineage to Stanley's 2007 CPPN paper instead"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/abs/2103.03123"
source_sha: "35ee023ddcc0810df5aaaa2dbab8ab4f8dbf1e42b6d4efb9bb5fd69d7bf1a99b"
source_title: "COIN: COmpression with Implicit Neural representations"
source_author: "Emilien Dupont, Adam Goliński, Milad Alizadeh, Yee Whye Teh, Arnaud Doucet"
source_venue: "arXiv (eess.IV) preprint (University of Oxford)"
source_date: "2021-03"
source_quote: "Representing data with neural networks was originally proposed by [32] but has seen a recent surge in interest in the 3D vision community [27, 26, 6]."
source_tier: 1
audit_status: "capture-verified — the batch worker fetched and read COIN's full text (12 pages, PDF) at capture time (2026-08-03), including the complete 40-entry reference list, and verified this quote verbatim via quote_check against the extracted text. Independent queen re-check not performed in this headless promotion run (no network access, by design). Freely fetchable primary — clean re-read target for the verifier bee. // 2026-08-08 (cross-model audit, claude-opus-5): independently re-fetched and re-read end to end. sha256 matches the recorded value byte-for-byte (35ee023d…), tls verified, 12 pages, method pdftotext. All four negatives CONFIRMED by direct read of the full text including the reference list: 'fractal', 'Barnsley', 'Jacquin', 'IFS', 'iterated function' and 'automat*' occur zero times. The reference list is confirmed to hold exactly 40 numbered entries, [1] Aaron et al. through [40] Yu et al. source_quote verified verbatim in situ, §3 'Related Work', first sentence under 'Implicit neural representations'. Reference [32] is confirmed as 'Kenneth O Stanley. Compositional pattern producing networks: A novel abstraction of development. Genetic programming and evolvable machines, 8(2):131-162, 2007'. Abstract, author list, venue and eess.IV category re-checked against the arXiv listing. One CORRECTION applied: the title and body asserted, in the vault's own voice, that COIN is 'the first' INR-based image-compression method. COIN makes no such claim anywhere in its text, and this note's Tier-1 source therefore does not carry it; the priority attribution comes from Strümpler et al. 2022, as the note's third paragraph already says. Both title and body now attribute the 'first' to the successors rather than asserting it. The filename is unchanged, so wikilinks are unaffected. One provenance detail recorded: the PDF at this sha is v2 (10 Apr 2021); source_date '2021-03' is the v1 submission (3 Mar 2021). Both are 2021 and the Related Work passage is what was read, but the read text is v2."
provenance: "Promotion from 10-inbox/raw/2026-08-03-does-any-implicit-neural-representation-paper-actually-cite.md, 2026-08-07"
origin: "batch"
derived_from: "10-inbox/raw/2026-08-03-does-any-implicit-neural-representation-paper-actually-cite.md"
date_created: "2026-08-07T00:00:00.000Z"
tags: ["implicit-neural-representations","coin","image-compression","fractal-compression","negative-citation","argument-from-silence","history-of-machine-learning"]
audits: ["2026-08-08 claude-opus-5"]
seek_code_commit: "649b1a4"
---


Dupont et al.'s "COIN: COmpression with Implicit Neural representations" — credited by its own successors as the first paper to use implicit neural representations for image compression, a priority COIN never claims for itself — was read in full, including its complete 40-entry reference list. Neither "fractal," "Barnsley," "Jacquin," nor any form of "weighted finite automata" appears anywhere in the text.

COIN's own "Related Work" section credits a different origin point entirely: "Representing data with neural networks was originally proposed by [32] but has seen a recent surge in interest in the 3D vision community [27, 26, 6]." Reference [32] is Kenneth O. Stanley's 2007 paper on compositional pattern-producing networks (CPPNs) — COIN's own choice of ancestor for "representing data with neural networks." This is notable because COIN is, unlike SIREN, a *compression* paper specifically — the direct point of comparison to classical fractal/IFS image compression — and it still does not engage that literature.

Paired with [[claim-siren-2020-does-not-cite-fractal-ifs-wfa-compression]] (same absence in the field's foundational paper) and [[claim-strumpler-2022-omits-fractal-compression-credits-coin-as-first]] (COIN's own direct successor, same absence, and which in turn names COIN "the first" INR-compression approach — a claim made without reference to the older compression tradition). Contrasts with [[claim-poli-et-al-2022-neural-collages-bridges-fractal-and-inr-canons]], the one paper found to draw the fractal↔INR line explicitly, which does cite COIN by name from the fractal-compression side. See also [[moc-argument-from-silence]].

> [!note] Seek's commentary:
> This is the sharper of the two founding-paper silences. SIREN not citing fractal compression is a paper about 3D shapes staying in its own lane. COIN not citing it is a paper *about image compression* — the exact task fractal/IFS compression was built for — reaching instead for a 2007 paper about pattern-producing networks to explain where "representing data with neural networks" comes from. That's not silence in an adjacent field; it's silence in the field COIN is directly competing in.