THE REGISTRAR'S COPY · WHAT THIS PROGRAM IS
The Classification

This is the program's identity, filed the way a registrar would file it. From a distance it will read as machine learning applied to geometry; up close the dependency runs the other way. It is the honed combination of computational geometry, information theory, machine learning, information dynamics, and dynamical systems — a Machine Learning / Artificial Intelligence research program built on a Programming-Languages-style geometric substrate, with the mathematics supplying the invariants: state, cost, dynamics, and global realizability. Every class below is earned by a named artifact in the record, not by affinity.

CROSS-LISTS · ARCHITECTURAL ORDER · EACH EARNED BY A NAMED ARTIFACT
Full nameCodeEarns its place by
Computational Geometry / Graphics Medial axis, frame fields, quad topology, the modeler application
Algebraic Topology Betti/Euler admission, Morse–Smale and Neumann domains, persistence-as-sensor
Spectral Theory Laplacian spectra, operator families, spectral certificates
Symbolic Computation The operator registry, symbolic regression over a typed algebra, grammar growth
Optimization and Control / Analysis of PDEs Transport-as-optimization, proximal evolution; optimal transport straddles both
Differential Geometry Finsler/Randers constitutive metrics, curvature-set transport
Dynamical Systems Koopman operators, generators, admissible evolution
Mathematical Physics The GENERIC/metriplectic formalism
Algebraic Geometry Divisors, meromorphic quartic differentials, Abel–Jacobi realizability
Functional Analysis The norm the gate must use: Banach completeness versus Hilbert projection, reproducing kernels and the Moore–Aronszajn constant, and the proof that an L2 residual bounds no supremum without a declared hypothesis
Numerical Analysis The discrete certificate: cotangent Laplacian and the discrete maximum principle, mass-matrix amplification, maximum-norm finite-element estimates, and the exactness of the nodal supremum for piecewise-linear output
Operator Algebras / High Energy Physics – Theory Structure only, and marked as such. Spectral triples and the gluing pairing supply the shape of a composition law — not a bound. The Atiyah state space of the relevant theory is one-dimensional, so nothing analytic crosses into the gates
THE THREE THAT NAME THE PROGRAM
PRIMARY

Machine Learning

The measured constitutive message-passing result, operator-reachability generalization, learned metrics and triage under an oracle.
CO-PRIMARY

Artificial Intelligence

The neurosymbolic architecture proper: staged synthesis, typed residuals, the promotion ratchet.
ARCHITECTURAL SUBSTRATE

Programming Languages

Not merely a cross-list. Typed intermediate representations, execution semantics, certificate propagation in the proof-carrying-code lineage.
The Construction Graph is not a representation for a model; it is the language in which geometry is executable.
READ IN ARCHITECTURAL ORDER, THE CLASSIFICATION BECOMES A SENTENCE
Machine Learning is the learned capability
→
Artificial Intelligence is the reasoning architecture
→
Programming Languages is the constitution
→
Computational Geometry is the domain
→
mathematics supplies the invariants
MATHEMATICS SUBJECT CLASSIFICATION 2020 · NAMES RESOLVED
CodeFull name
Artificial neural networks and deep learning
Logic in artificial intelligence
Theory of compilers and interpretersTHE DISTINCTIVE ONE
Differentials on Riemann surfaces
Jacobians and Prym varieties
Persistent homology and applications, topological data analysis
Spectral problems; spectral geometry
Local differential geometry of Finsler spaces
Optimal transportation
Optimization of shapes
Numerical treatment of dynamical systems
Irreversible thermodynamics (the GENERIC home)
Hilbert spaces with reproducing kernels (the sharp sup-norm constant)
Sobolev spaces and embedding theorems (the s > d/2 threshold)
Asymptotic distribution of eigenvalues (the local Weyl law)
Finite element methods for boundary value problems
Noncommutative geometry — structure only
Topological field theories — structure only
EACH MATHEMATICAL DISCIPLINE ANSWERS ONE QUESTION THE MACHINE MUST ASK
Topology
What structure can exist?
Spectral theory
How is that structure represented?
Transport
How do two structures correspond?
Constitutive physics
What transformations are admissible, and at what cost?
Algebraic geometry
Which local proposals can coexist globally?
Functional analysis
In which norm is the claim true — and does that norm discharge the obligation?
Compiler theory
Which programs are legal?

And machine learning asks the sixth question — which admissible construction is worth proposing next — which is why it is primary without being sovereign.

This program will read, from a distance, as “machine learning applied to geometry.” Up close the dependency runs the other way: the deepest object in the architecture is the typed state-transition language connecting every layer; the mathematics gives that language its notions of state, cost, dynamics, and global realizability; the compiler gives it syntax, semantics, and execution; and learning is the mechanism by which the language grows. The predominant academic lineage is therefore a Machine Learning / Artificial Intelligence research program built on a Programming-Languages-style geometric substrate — and the substrate, not the network, is what the other disciplines are holding up.

This page mirrors §4 of The Orientation Atlas, where it lives beside the architecture. The classification runs as an instrument twice over: Figure 03 draws its two axes, and the Compiler Atlas (press ⌖ there) applies it stage by stage to any proposed ML architecture.