Does Yaser Abu-Mostafa's own learning-theory work cite any optimal-control / LQG source directly?

yaser-abu-mostafanipsneuripsvc-dimensionlearning-theoryoptimal-controllqgdisciplinary-wallcitation-analysisvapnik-chervonenkis

Short answer: no, not in any of the three primary theoretical papers read in full this session. Across Abu-Mostafa's foundational 1989 VC-dimension paper, his 1992/93 "A Method for Learning from Hints," and his 1995 "Hints" review (his most comprehensive synthesis of the VC-dimension/hints theory line, with a 39-entry reference list), zero citations to Kalman, Bellman, Pontryagin, Riccati, Bryson, Bryson-Ho, Kwakernaak, Kwakernaak-Sivan, "optimal control," "LQG," or "dynamic programming" appear. The one optimization-adjacent citation across all three reference lists — Wismer and Chattergy's Introduction to Nonlinear Optimization (1978) — is a nonlinear-programming text, not a control-theory one. This answers question-verify-abu-mostafa-learning-theory-cites-optimal-control-lqg from the strongest angle available (primary reference-list reads), strengthening rather than overturning observation-nips-founders-kwakernaak-twente-founding-cosine-pairing-is-embedding-false-friend's false-friend verdict. The gap: his 2012 textbook Learning From Data (with Magdon-Ismail and Lin) was not obtained and read this session, so the check is not exhaustive over his full career output.

Claim: Abu-Mostafa's 1995 "Hints" review (Neural Computation) — his most comprehensive theoretical synthesis — cites no optimal-control or LQG source across its full reference list

verifies: question-verify-abu-mostafa-learning-theory-cites-optimal-control-lqg

• Claim type: historical/bibliographic (what a specific document's reference list contains) — load-bearing for the vault's question, so escalated to the Tier 1–2 floor. Met: primary document read in full via extract_pdf, direct fetch, TLS verified.
• source_url: https://home.work.caltech.edu/pub/Abu-Mostafa1995hints.pdf
• source_author: Yaser S. Abu-Mostafa
• source_date: 1995
• source_title: "Hints"
• source_venue: Neural Computation 7, 639–671, MIT Press
• source_sha: c865995f82bac96c91a54cf7097ed6b83571b2a87684a14573f4ddce9f688803 (tls: verified; method: ocr)
• source_tier: 1 — author's own paper, own institutional-site PDF, read directly and completely (33 pages, abstract through references)
• source_quote (opening framing of the paper's actual mathematical lineage): "The use of hints is tantamount to combining rules and data in learning, and is compatible with different learning models, optimization techniques, and regularization techniques... On the theoretical side, the information value of hints is contrasted to the complexity value and related to the VC dimension."
• source_quote (the full References section was read end to end, alphabetical Abu-Mostafa through Wismer/Chattergy — 26 entries): entries include Akaike (1969, autoregressive models), Blumer/Ehrenfeucht/Haussler/Warmuth (1989, VC-dimension learnability), Cover and Thomas (1991, information theory), Duda and Hart (1973, pattern classification), Hertz/Krogh/Palmer (1991, neural computation theory), Malkiel (1973, A Random Walk Down Wall Street), Minsky and Papert (1988, Perceptrons), Rumelhart/Hinton/Williams (1986, backpropagation), Vapnik and Chervonenkis (1971), and "Wismer, D., and Chattergy, R. 1978. Introduction to Nonlinear Optimization. North Holland, Amsterdam." No entry cites Kalman, Bellman, Pontryagin, Riccati, Bryson, Bryson-Ho, Kwakernaak, Kwakernaak-Sivan, or any work with "control" in the title.
• The paper's own stated theoretical ancestor is explicit and singular: Vapnik and Chervonenkis (1971), cited as the source of "the key tools for dealing with the information issue" (per the companion 1989 paper below) — statistical learning theory, not control theory.

Claim: Abu-Mostafa's foundational 1989 VC-dimension paper and 1992/93 NIPS hints paper independently confirm the same absence

verifies: question-verify-abu-mostafa-learning-theory-cites-optimal-control-lqg

• Claim type: historical/bibliographic, same floor and same method as above (both read in full via extract_pdf, direct fetch, TLS verified).
• Source A — 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_sha: 0bdc18282f2d0dbc1b3964da0e68fc2f396847354b8a0e1cf74f5ef82b3afe95 (tls: verified); source_tier: 1. Full reference list (6 entries, read verbatim): Baum and Haussler (1989), Blumer/Ehrenfeucht/Haussler/Warmuth (1986), Cover (1965), Judd (1988), Rumelhart/Hinton/Williams (1986), and "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." No control-theory citation of any kind.
• Source B — source_url: https://authors.library.caltech.edu/records/mwmvb-va834/files/616-a-method-for-learning-from-hints.pdf; source_author: Yaser S. Abu-Mostafa; source_date: 1993 (NIPS 1992 conference, published as Advances in Neural Information Processing Systems 5, pp. 73–80); source_title: "A Method for Learning from Hints"; source_venue: Morgan Kaufmann / NIPS proceedings; source_sha: 4d49d96ec0ebd09417befb65ef6e6ba332a8d265515c1fd725988d2ad1da700f (tls: verified); source_tier: 1. Full reference list (6 entries, read verbatim): "Abu-Mostafa, Y. S. (1990)"; "Al-Mashouq, K. and Reed, I. (1991)"; "Minsky, M. L. and Papert, S. A. (1969)"; "Omlin, C. and Giles, C. L. (1992)"; "Suddarth, S. and Holden, A. (1991)"; and "Wismer, D. A. and Chattergy, R. (1978), 'Introduction to Nonlinear Optimization,' North Holland." This last is the paper's own citation for its "penalty function" approach to balancing multiple error terms — a nonlinear-programming reference, not an optimal-control one; no Kalman/Bellman/Pontryagin/Riccati/Bryson/LQG citation appears.
• Together with the 1995 "Hints" review, this covers three of Abu-Mostafa's core theoretical statements on VC-dimension and hints (1989, 1992/93, 1995) — the primary line of his own learning-theory contribution — with a consistent negative result across all three independently-alphabetized reference lists.

Claim: Abu-Mostafa's complete self-maintained publications list shows no title referencing optimal control, LQG, or Kalman filtering across ~48 works (1984–2026)

verifies: question-verify-abu-mostafa-learning-theory-cites-optimal-control-lqg

• Claim type: historical/bibliographic, weaker/supporting evidence (title-level scan, not a reference-list read of every paper) — Tier 1 source (his own faculty page) but the claim itself is only about titles, so it corroborates rather than independently proves the absence inside reference lists.
• source_url: https://work.caltech.edu/pubs.html
• source_author: Yaser S. Abu-Mostafa (self-maintained faculty page)
• source_date: page undated; publications listed span September 1984 to August 2026
• source_title: "Yaser S. Abu-Mostafa — Publications" (page header: "P U B L I C A T I O N S")
• source_venue: Caltech (author's own institutional site, work.caltech.edu)
• source_sha: cf17eb8a15ea4541e9ca19295fac95326c7e7951c1bdd6d37eebb9e66f36950c
• source_tier: 1 — the author's own maintained record of his career output, fetched directly
• source_quote (representative titles spanning the theoretical and applied range): "The Vapnik-Chervonenkis Dimension: Information versus Complexity in Learning," "Hints," "Financial Model Calibration Using Consistency Hints," "Machine Learning for Recession Prediction and Dynamic Asset Allocation," "Learning From Data," "Optical neural computers," "Information Capacity of the Hopfield Model" — 48 listed works total, none titled around optimal control, LQG, Kalman filtering, Pontryagin's principle, or the Riccati equation.
• Caveat: absence of a title is weaker evidence than absence-in-a-reference-list (a paper can cite a field without naming it in its own title); this claim is recorded as corroborating context for the two claims above, not as a standalone proof.

Safety flags

None fired. All fetched pages (four PDFs, one HTML publications page) were plain academic/bibliographic content — abstracts, equations, reference lists, a faculty publication listing — with no addressed-to-AI language, override language, claimed authority, tier self-assignment, file-system instructions, credential requests, or urgency framing. All PDF fetches via extract_pdf reported tls: verified.

Further leads

• Obtain and read the reference list of Learning From Data (Abu-Mostafa, Magdon-Ismail, and Lin, AMLbook.com, 2012) — his most-cited and most comprehensive theory text — which was not located as an open-access PDF this session; this is the single biggest remaining gap in a full-career check.
• Check "Financial Model Calibration Using Consistency Hints" (IEEE Trans. Neural Networks, 2001) and the financial-forecasting papers (with A. Atiya) for possible incidental citations to stochastic control or filtering literature — the financial-applications line is the part of his output most likely, a priori, to brush against control-theoretic filtering (e.g., Kalman filters are common in quantitative finance), and was not checked this session.
• "Monotonicity Hints" (Sill and Abu-Mostafa, NIPS 1996) and "Financial Applications of Learning from Hints" (NIPS 1995) reference lists were not read this session; both are readily available from the same work.caltech.edu / CaltechAUTHORS hosts.
• The Caltech "Learning Systems Group" page and Abu-Mostafa's course materials (the Caltech "Learning From Data" MOOC lecture slides) may cite control theory in passing even if the published papers do not; not checked.

Entity candidates

• Vladimir Vapnik — person — with Chervonenkis, the actual foundational figure Abu-Mostafa's entire theoretical program measures itself against; cited in every one of the three papers read here as the origin of the VC-dimension framework. This is the true intellectual ancestry the question implicitly asks about, and it runs to statistical learning theory, not control theory.
• Alexey Chervonenkis — person — co-author of the 1971 paper that is the singular cited theoretical ancestor across all three Abu-Mostafa papers checked
• Yaser Abu-Mostafa — person — subject of this capture; already has entity-yaser-abu-mostafa
• Malik Magdon-Ismail — person — Abu-Mostafa's most frequent co-author across the financial-learning and Learning From Data line; textbook reference list (unread this session) would be his to check too
• Hsuan-Tien Lin — person — third co-author of Learning From Data
• David E. Rumelhart — person — cited (with Hinton and Williams, 1986 backpropagation paper) in both the 1989 VC-dimension paper and the 1995 Hints review; the connectionist-side citation these papers do make, contrasted against the control-theory citation they do not make
• Wismer and Chattergy — persons (D. Wismer, R. Chattergy) — authors of the one "optimization" (not "optimal control") citation shared across two of the three papers; worth noting explicitly as the closest the reference lists come to the control/optimization literature, and how far that still is from LQG
• Learning From Data (book) — term/document — the 2012 textbook whose bibliography remains the one major unchecked primary source for this question