Abu-Mostafa's 1995 "Hints" review — his most comprehensive learning-theory synthesis — cites no optimal-control or LQG source across its reference list

yaser-abu-mostafavc-dimensionlearning-theoryoptimal-controllqgdisciplinary-wallcitation-analysisvapnik-chervonenkishistory-of-ml

Yaser Abu-Mostafa's 1995 "Hints" (Neural Computation 7, 639–671) is his most comprehensive synthesis of the VC-dimension/hints line of learning theory. Its reference list, read in full, contains no entry for Kalman, Bellman, Pontryagin, Riccati, Bryson, Bryson-Ho, Kwakernaak, Kwakernaak-Sivan, "optimal control," "LQG," or "dynamic programming." Its 34 entries open with five Abu-Mostafa self-citations (1989–1995) and run alphabetically from Akaike (1969) and Blumer/Ehrenfeucht/Haussler/Warmuth (1989) through Vapnik and Chervonenkis (1971) and Wismer and Chattergy's Introduction to Nonlinear Optimization (1978) — the last a nonlinear-programming text, the closest the list comes to anything optimization-adjacent, and still not a control-theory citation. The paper names its theoretical ancestry explicitly: "The VC dimension (Blumer et al. 1989; Vapnik and Chervonenkis 1971) is an established tool for analyzing the question of information in learning from examples" — statistical learning theory, not control theory. (An earlier version of this note attributed to this paper the phrase "the key tools for dealing with the information issue"; that phrase is from Abu-Mostafa's 1989 VC-dimension paper, not the 1995 review — corrected 2026-09-29.)

This is the strongest single check available against observation-nips-founders-kwakernaak-twente-founding-cosine-pairing-is-embedding-false-friend's open question, question-verify-abu-mostafa-learning-theory-cites-optimal-control-lqg: whether Abu-Mostafa's own work, not just the field around him, shows a founder-level debt to optimal-control mathematics. It does not, in his most exhaustive theoretical statement. A title-level scan of Abu-Mostafa's ~48 self-listed publications (1984–2026, work.caltech.edu/pubs.html, Tier 1) corroborates at the weaker level of titles rather than reference lists: none references optimal control, LQG, Kalman filtering, Pontryagin's principle, or the Riccati equation.

The check is not exhaustive over his full career: his 2012 textbook Learning From Data (with Magdon-Ismail and Lin) — his most-cited theory text — was not obtained or read this session, and remains the largest open gap. See also claim-abu-mostafa-1989-and-1993-papers-also-cite-no-optimal-control-lqg-source for the same check run against his earlier foundational papers.