The frontier map — territory 4 of 5

Optimization & Game Theory

The engineer runs optimizers; the scientist knows when and why they converge — and what breaks when two networks optimize against each other. This territory spans convex analysis, convergence rates, minimax dynamics and optimal transport: the mathematics behind Adam, WGAN and PPO.

The central picture: three landscapes, three destinies

What the loss surface allows determines what theory can promise

Convex — solved one global minimum provable rates: 1/t, e^−t classical ML, logistic regression Non-convex — deep learning many valleys & saddles few guarantees — yet SGD finds good minima anyway (open theory) Minimax — games (GANs, RL) two players, one saddle point simultaneous descent can orbit forever — why GAN training oscillates
Deep learning lives in the middle panel with the frontier still open; GANs and RL live in the right panel, where "just run gradient descent" provably isn't enough.

The roadmap of understanding — three levels deep

Level 1 — Core objects

the shapes of solvability
convexity  ·  local vs global minima  ·  convergence rates

The ideas

  • Convexity: every local minimum is global — the "easy" world
  • Rates: how fast error shrinks (1/t vs 1/t² vs e^−t)
  • Saddle points and plateaus — the honest hard parts

Where it lives

  • Classic ML's guarantees; deep learning's lack of them

What it unlocks

  • Knowing which promises theory can and cannot make

Level 2 — Working theory

the machinery behind every optimizer
smoothness & strong convexity  ·  acceleration  ·  adaptive methods  ·  proximal & constrained

The ideas

  • Smoothness L and strong convexity μ; condition number κ = L/μ
  • Momentum & Nesterov acceleration: provably faster, not a hack
  • Adaptive methods: Adam as diagonal preconditioning; why AdamW decouples weight decay
  • Proximal operators & constraints; large-batch scaling (LARS/LAMB)

Where it lives

  • Every optimizer paper's convergence section
  • LR-scaling rules when batch size changes
  • Second-order-flavored optimizers (Shampoo, Muon) at LLM scale

What it unlocks

  • Designing optimizers and schedules from principles
  • Diagnosing "slow training" as conditioning, not bad luck

Level 3 — The frontier

games, transport, and nested problems
minimax dynamics  ·  optimal transport  ·  trust regions  ·  bilevel optimization

The ideas

  • Minimax optimization: Nash equilibria, rotational dynamics, extragradient fixes
  • Optimal transport: Wasserstein distance as the geometry between distributions
  • Trust regions: constrain each step's damage (TRPO → PPO's clipping)
  • Bilevel problems: optimize through an optimization (meta-learning, RLHF)

Where it lives

  • The WGAN story in full: vanishing gradients weren't a bug in the network — the JS divergence gave no useful slope between disjoint distributions; Wasserstein distance restored the geometry of descent
  • PPO — the optimizer of RLHF
  • Multi-agent learning, self-play (AlphaZero-style)

What it unlocks

  • Stabilizing any adversarial or RL training scheme
  • Choosing the distance a model minimizes — often the real design decision
  • Reading alignment methods as constrained optimization

Cheat sheet: concept → landmark → what it's doing there

ConceptLandmark useWhat it's doing there
Acceleration (momentum)Nesterov; every training recipeProvably faster descent on ill-conditioned bowls
Diagonal preconditioningAdam / AdamWA cheap approximation of curvature, per weight
Minimax dynamicsGAN training theoryWhy simultaneous descent orbits; how to damp it
Optimal transport / WassersteinWGANA distance with usable gradients between distributions
Trust regionsTRPO → PPO (RLHF)Bounding each policy update's harm
Bilevel optimizationMeta-learning, hyperparameter searchLearning through an inner learner
The territory's signature move: when training misbehaves structurally, change the problem, not the hyperparameters — a better distance (WGAN), a constrained step (PPO), an averaged iterate. The deepest fixes in this field are reformulations.
← Territory 3: Learning TheoryGeneralization & scaling laws