Course Catalog & Credential Pathway — Dwg. No. PF-002D
A field branch of the Construction Track, for students aimed at real-time plasma control — reinforcement learning applied to a reactor instead of a game.
Field branch of Dwg. No. PF-002, The Construction Track — same core sequence through Phase II, concentrated here into plasma physics & reactor control from Phase II onward. See also Dwg. No. PF-001, The Verification Track.
Five core strands, plus the domain strand this branch concentrates in
No frameworks yet, and no plasma physics yet either. Two years of straight math and programming fundamentals, identical to the core Construction Track.
Precalculus, Accelerated
Compressed to clear room for calculus by grade 11 — the standard on-ramp for anyone headed toward a quantitative degree.
Programming I — Python
Variables, control flow, functions, data structures. Fluency, not tricks.
Discrete Mathematics & Logic
Sets, proofs, graphs, combinatorics — the math CS theory actually runs on, usually skipped until it's overdue.
Technical Writing for Research
Writing a clear methods section and an honest results section — the two paragraphs every paper lives or dies on.
The math and CS sequences converge into an actual first model, same as the core track. The plasma-physics layer begins concurrently — by the end of Phase II a student can both train a small RL policy and read the MHD equilibrium equations a tokamak controller has to respect.
Calculus I & II
Through multivariable and the gradient — backpropagation is the chain rule with bookkeeping, and it should read that way.
MATH 101
Linear Algebra
Vector spaces, eigendecomposition, matrix calculus. Every tensor operation is this course wearing a framework's syntax.
MATH 101
Probability & Statistics for ML
Distributions, estimation, Bayes — framed toward loss functions and uncertainty, not toward the social-science stats track.
Data Structures & Algorithms
Complexity analysis and the standard structures, drilled to fluency — still the baseline technical-interview bar at every AI lab.
CS 100
Intro to Machine Learning
Regression through a first neural net, built from array operations before any framework is allowed to hide the mechanics.
MATH 210, STAT 220
Research Seminar — Reading Group
Weekly seminal-paper reads (perceptron through transformers), presented and defended aloud, not just summarized.
Plasma Physics Fundamentals
Charged-particle motion, plasma confinement, and the instabilities that make a tokamak's plasma so hard to hold in shape.
MATH 210 (concurrent)
Magnetohydrodynamics & Tokamak Control
MHD equilibrium, magnetic-coil actuation, and the classical control theory a plasma controller has to outperform to be worth deploying.
PHYS 230
Here the domain degree takes over from the general CS/applied-math track. A nuclear or plasma physics-plus-CS program replaces DEG 300 — fusion labs hire for reactor physics first, control engineering second.
Nuclear / Plasma Physics Degree
Formal core in a nuclear or plasma physics-plus-CS program — the credential fusion labs hire against ahead of control-engineering skill alone.
Reinforcement Learning for Plasma Control
Simulation-trained RL control policies for real-time plasma shaping, in the DeepMind–EPFL tradition — and the sim-to-real gap that decides whether a policy survives contact with an actual reactor.
FUS 240, ML 230
Deep Reinforcement Learning
Policy-gradient and actor-critic methods in depth, taught directly against real-time control problems — tokamak plasma shaping among them — rather than as a generic survey.
ML 230
Systems for ML
Distributed training, GPU/accelerator programming, the infrastructure that turns a notebook model into one that trains and runs in real time — the same infrastructure a reactor control loop needs.
CS 210
Supervised Research Practicum
Placement inside a fusion lab or plasma-control research group, with a mentor of record. Graded on a reproduction that actually reproduces, or a contribution that gets merged.
RES 240, ML 310
Capstone — Build & Ship
One model or tool — a control-policy benchmark suite, a disruption-prediction model, a simulation-to-real transfer study — taken from idea to a deployed, load-bearing artifact with real users.
ENG 320, RES 350
No end date: the ML baseline moves every conference cycle, and the reactor-physics baseline — disruption prediction, confinement records, new tokamak and stellarator designs — moves on its own slower schedule that a purely computational graduate can lose track of.