\( \newcommand{\vect}[1]{\mathbf{#1}} \newcommand{\mat}[1]{\mathbf{#1}} \newcommand{\R}{\mathbb{R}} \newcommand{\E}{\mathbb{E}} \newcommand{\Prob}{\mathbb{P}} \newcommand{\T}{^{\top}} \newcommand{\norm}[1]{\lVert #1 \rVert} \newcommand{\abs}[1]{\lvert #1 \rvert} \newcommand{\ihat}{\mkern1mu\hat{\boldsymbol{\imath}}\mkern2mu} \newcommand{\jhat}{\mkern1mu\hat{\boldsymbol{\jmath}}\mkern2mu} \newcommand{\pd}[2]{\frac{\partial #1}{\partial #2}} \newcommand{\argmax}{\operatorname*{arg\,max}} \newcommand{\argmin}{\operatorname*{arg\,min}} \newcommand{\cdata}[1]{\class{c-data}{#1}} \newcommand{\cmodel}[1]{\class{c-model}{#1}} \newcommand{\ctruth}[1]{\class{c-truth}{#1}} \newcommand{\cgrad}[1]{\class{c-grad}{#1}} \newcommand{\closs}[1]{\class{c-loss}{#1}} \newcommand{\cparam}[1]{\class{c-param}{#1}} \newcommand{\csecond}[1]{\class{c-second}{#1}} \renewcommand{\,}{\mkern3mu} \)

RL03 · Deep Value-Based RL

A free, illustrated L3 course of 8 modules in the Reinforcement Learning track. Modules include: Function approximation and the deadly triad; Deep Q-networks; Experience replay and target networks.

RL03 L3 · Advanced 8 modules · about 32 lessons Planned · Ring 3

Part of the Reinforcement Learning track. Ships in Ring 3 (Breadth) of the roadmap.

Modules

1. Function approximation and the deadly triad
2. Deep Q-networks
3. Experience replay and target networks
4. Double, dueling and prioritized variants
5. Distributional RL
6. Rainbow and its ablations
7. Q-functions for continuous actions
8. Tuning and debugging in practice

Signature illustrations

The pictures a learner should still see in their head a year later:

Data flow through a deep Q-network, sampling from a replay buffer, overestimation bias plots, histograms of return distributions.

Status

This course is planned, not written yet. It ships with Ring 3 (Breadth).

Every lesson will follow the same twelve-part blueprint with at least three purpose-made figures. Watch the repository to hear when modules ship, or help write and review it.