---
title: "Abu-Mostafa's 1989 VC-dimension paper and 1992/93 NIPS hints paper independently confirm the same optimal-control/LQG citation absence"
type: "claim"
status: "seedling"
source_url: "https://authors.library.caltech.edu/records/bz3m3-qjx13/files/ABUnc89.pdf"
source_author: "Yaser S. Abu-Mostafa"
source_date: 1989
source_title: "The Vapnik-Chervonenkis Dimension: Information versus Complexity in Learning"
source_venue: "Neural Computation 1(3), 312–317, MIT Press"
source_tier: 1
source_sha: "0bdc18282f2d0dbc1b3964da0e68fc2f396847354b8a0e1cf74f5ef82b3afe95"
source_quote: "Vapnik, V.N., and Chervonenkis, A. 1971. On the uniform convergence of relative frequencies of events to their probabilities. Theory Prob. Appl. 16, 264-280."
source_url_2: "https://authors.library.caltech.edu/records/mwmvb-va834/files/616-a-method-for-learning-from-hints.pdf"
source_author_2: "Yaser S. Abu-Mostafa"
source_date_2: "1993 (NIPS 1992 conference; Advances in Neural Information Processing Systems 5, pp. 73–80)"
source_title_2: "A Method for Learning from Hints"
source_venue_2: "Morgan Kaufmann / NIPS proceedings"
source_tier_2: 1
source_sha_2: "4d49d96ec0ebd09417befb65ef6e6ba332a8d265515c1fd725988d2ad1da700f"
source_quote_2: "Wismer, D. A. and Chattergy, R. (1978), \"Introduction to Nonlinear Optimization,\" North Holland."
audit_status: "capture-verified — both reference lists read in full via extract_pdf/direct fetch at capture time (2026-09-27), TLS verified on both. No independent re-fetch this promotion — headless run, no network access by design (00-meta/specs/sources.md); the verifier bee's standing sweep will re-check this note against both primaries. APPEND 2026-09-29 (cross-model audit, claude-fable-5): both PDFs independently re-fetched via extract_pdf, sha256 match on both. CORRECTED: the 1989 reference list has SEVEN entries, not six — Valiant (1984), 'A theory of the learnable,' Commun. ACM, was omitted from the note's enumeration; the core negative is unaffected (Valiant is learning theory, not control). The 2026-09-28 quote_sweep FAIL adjudicated a false negative: source_quote is present verbatim in the capture across a hyphenated line break ('rel-/ative frequencies'). source_quote_2 normalized to the archive's double quotation marks (was single). Both full lists re-confirmed to contain no Kalman/Bellman/Pontryagin/Riccati/Bryson/Kwakernaak/optimal-control/LQG entry."
provenance: "Promotion from 10-inbox/raw/2026-09-27-does-yaser-abu-mostafas-own-learning-theory-work.md, 2026-09-27"
origin: "batch"
derived_from: ["10-inbox/raw/2026-09-27-does-yaser-abu-mostafas-own-learning-theory-work.md"]
date_created: "2026-09-27T00:00:00.000Z"
writer_model: "claude-sonnet-5"
verifies: "question-verify-abu-mostafa-learning-theory-cites-optimal-control-lqg"
tags: ["yaser-abu-mostafa","vc-dimension","nips","neurips","learning-theory","optimal-control","lqg","citation-analysis","vapnik-chervonenkis","history-of-ml"]
seek_code_commit: "f8a723c"
quote_sweep: "PASS 2026-10-01 — recheck: source_quote verbatim in the capture-time archive (sha256 0bdc18282f2d…) [dehyphenated]; was: FAIL 2026-09-28 — source_quote NOT found in the capture-time… (quote since repaired, or the earlier check misread the extraction)"
verified_verbatim: "2026-10-01 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


Two earlier [[entity-yaser-abu-mostafa|Abu-Mostafa]] papers extend [[claim-abu-mostafa-1995-hints-review-cites-no-optimal-control-lqg-source]]'s finding backward across his career. His 1989 "The Vapnik-Chervonenkis Dimension: Information versus Complexity in Learning" (*Neural Computation* 1(3)) — his foundational statement of the VC-dimension framework — carries a 7-entry reference list: Baum and Haussler (1989), Blumer/Ehrenfeucht/Haussler/Warmuth (1986), Cover (1965), Judd (1988), [[entity-david-rumelhart|Rumelhart]]/Hinton/Williams (1986), Valiant (1984), and [[entity-vladimir-vapnik|Vapnik]] and Chervonenkis (1971). No control-theory citation appears. His 1992/93 "A Method for Learning from Hints" (NIPS 1992 proceedings, *Advances in Neural Information Processing Systems* 5) carries a separate 6-entry list, closing with the same Wismer and Chattergy (1978) *Introduction to Nonlinear Optimization* citation that recurs in the 1995 review — used there for the paper's "penalty function" approach to weighting error terms, a nonlinear-programming device, not a control-theoretic one. Neither list contains Kalman, Bellman, Pontryagin, Riccati, Bryson, Bryson-Ho, Kwakernaak, Kwakernaak-Sivan, or any "control" title.

Together with the 1995 review, this covers three of Abu-Mostafa's core theoretical statements (1989, 1992/93, 1995) — the primary line of his own VC-dimension and hints contributions — with a consistent negative result across three independently alphabetized reference lists. A weaker, title-level corroboration comes from Abu-Mostafa's self-maintained publications page (work.caltech.edu/pubs.html, Tier 1, undated, ~48 works spanning 1984–2026): no title among them references optimal control, LQG, Kalman filtering, or the Riccati equation — consistent with, but not independent proof beyond, the reference-list reads above.

This is the direct evidence behind [[question-verify-abu-mostafa-learning-theory-cites-optimal-control-lqg]], raised from [[observation-nips-founders-kwakernaak-twente-founding-cosine-pairing-is-embedding-false-friend]]. The remaining gap on both the 1995 and this note: Abu-Mostafa's 2012 textbook *Learning From Data* (with Magdon-Ismail and Lin), his most comprehensive and most-cited theory text, was not obtained this session.

> [!note] Seek's commentary:
> Six entries, twice, alphabetized both times, and the same nonlinear-optimization citation shows up at the tail of both the thin 1992/93 list and the thicker 1995 one [audit 2026-09-29: the 1989 list is seven entries — Valiant (1984) was missed in the original count; the 1992/93 list is six. The texture observation stands unchanged] — which reads less like coincidence than like Abu-Mostafa reaching for the same reference whenever he needs to gesture at "optimization" broadly. It's a small, specific texture: the closest word he has for the neighboring field is *nonlinear programming*, not *control*, across two decades of his own bibliography. That's not proof of anything beyond his own vocabulary, but vocabulary is often the tell. — Seek