NNeural Field Guide
0 / 32 mastered
Reinforcement Learning & Control · from first principles

Learn to decide.
Measure every consequence.

Start with action, consequence, and partial observation, then build value learning, policies, planning, safety, and research evidence. Formal equations follow the decision mechanism instead of leading it.

32 lessons · about 15 hours
THE CONTROL LOOP
observation→policy + action
consequencereward + next stateillustrative flow · not a measurement
Why Reinforcement Learning & Control

Teach systems to improve through consequences.

Reinforcement learning is where prediction becomes decision-making. It explains how an agent can value delayed outcomes, explore an uncertain world, improve from experience, and still respect constraints that reward alone cannot express.

01 · Reason

See the decision loop clearly

Separate state, action, reward, return, value, policy, and data so every update has an inspectable meaning.

02 · Build

Move from tables to deep agents

Implement value methods, DQN, actor-critic, planning, offline learning, and model-based control as connected systems.

03 · Evaluate

Know when an agent truly improved

Use matched budgets, multiple seeds, support checks, constraints, and failure evidence instead of trusting one impressive run.

The finish line

Finish with the tools to build learning agents—and the discipline to tell progress from noise.

What you will build

From one transition to an evidence-bounded learning agent.

Seven territories expose values, policies, models, logged behavior, constraints, and independent-run evidence as one cumulative control system.

NEURAL FIELD GUIDE / COURSE ARC01 · 07
01 / 01 · Lessons 1–5

Sequential Decisions

Specify and solve a small decision process.

You unlockTested MDP solver and rollout evaluator
02 / 02 · Lessons 6–10

Values & Temporal Difference

Implement Monte Carlo, TD, SARSA, and Q-learning.

You unlockChanged-environment value-method comparison
03 / 03 · Lessons 11–15

Deep Value Learning

Build and diagnose a DQN-style agent.

You unlockPinned DQN reproduction and failure dossier
04 / 04 · Lessons 16–19

Policy Gradients

Build policy-gradient and actor-critic agents.

You unlockPolicy-gradient/actor-critic comparison
05 / 05 · Lessons 20–24

Planning & Model-Based RL

Combine learned dynamics, search, and policy learning.

You unlockMatched-budget planner/policy study
06 / 06 · Lessons 25–28

Imitation & Offline RL

Diagnose coverage and build an offline policy pipeline.

You unlockDecision-sequence policy and coverage dossier
07 / 07 · Lessons 29–32

Reliable RL Research

Reproduce a baseline and run a seed-level intervention study.

You unlockSeed-level controlled RL study
Inside the course

One course. 7 connected frontiers.

Lessons 01–32 build working tabular, deep-value, actor-critic, model-based, and offline agents, then culminate in a seed-level original study.

01
Define the loop

Sequential Decisions

Specify and solve a small decision process. Each lesson extends one inspectable agent and evidence system.

The payoffSpecify and solve a small decision process.
5 lessons130 min
02
Back up outcomes

Values & Temporal Difference

Implement Monte Carlo, TD, SARSA, and Q-learning. Each lesson extends one inspectable agent and evidence system.

The payoffImplement Monte Carlo, TD, SARSA, and Q-learning.
5 lessons134 min
03
Stabilize approximation

Deep Value Learning

Build and diagnose a DQN-style agent. Each lesson extends one inspectable agent and evidence system.

The payoffBuild and diagnose a DQN-style agent.
5 lessons136 min
04
Differentiate behavior

Policy Gradients

Build policy-gradient and actor-critic agents. Each lesson extends one inspectable agent and evidence system.

The payoffBuild policy-gradient and actor-critic agents.
4 lessons112 min
05
Learn then imagine

Planning & Model-Based RL

Combine learned dynamics, search, and policy learning. Each lesson extends one inspectable agent and evidence system.

The payoffCombine learned dynamics, search, and policy learning.
5 lessons138 min
06
Learn from logged behavior

Imitation & Offline RL

Diagnose coverage and build an offline policy pipeline. Each lesson extends one inspectable agent and evidence system.

The payoffDiagnose coverage and build an offline policy pipeline.
4 lessons114 min
07
Measure policies honestly

Reliable RL Research

Reproduce a baseline and run a seed-level intervention study. Each lesson extends one inspectable agent and evidence system.

The payoffReproduce a baseline and run a seed-level intervention study.
4 lessons120 min
32connected lessons
32hands-on labs
32code notebooks

No black boxes. Build intuition, see the mechanism, then make the real engineering trade-offs.

Built from the field, not the hype

Primary work behind the course

Ready when you are

Finish with the tools to build learning agents—and the discipline to tell progress from noise.

32 connected lessons, hands-on labs, and a complete end-to-end build.