← all subjects

Reinforcement Learning

Most machine learning trains on labeled answers. Reinforcement learning gets only a number — a reward — and must work out, from its own trial and error, which of its actions earned it.

The setting is an agent acting in an environment: at each step it observes a state, picks an action, and receives a reward and a next state. The goal is not to match a label but to maximize reward accumulated over the long run — which means trading a sure gain now against a larger one later.

FIG_002
s0s_0
s1s_1
s2s_2
A Markov decision process: states, transitions, and the actions between.

One idea organizes the subject: learn a value, the expected long-run reward of a state, and act greedily toward it. The methods differ in how they estimate that value: sweeping a known model with dynamic programming, averaging sampled returns with Monte Carlo, or temporal-difference learning, which updates each estimate toward the next before the episode is even over.

FIG_003
G
A gridworld: value spreads from the goal and the greedy policy points home.

Bootstrapping from one estimate to the next is what makes it work online. It is also what the brain's dopamine neurons appear to compute: a reward-prediction error, the same signal the algorithms are built on.

FIG_001
AGENTENVIRONMENT
ata_t
st+1, rt+1s_{t+1},\ r_{t+1}
The loop: the agent acts, the environment answers with a state and reward.
FIG_004
s0s_0
s1s_1
s2s_2
s3s_3
s4s_4
reward
Temporal-difference learning: the prediction error backs up along the chain.

Scale the tables into function approximators and the same equations drive deep reinforcement learning: the Q-network that first played Atari from pixels, the policy gradients behind modern robot control, and the self-play search that took AlphaZero from the rules of Go to superhuman in a day.

FIG_005
12345
A bandit: pulls sharpen each arm's estimate until the best one wins out.

Contents.

·
Progress.░░░░░░░░░░░░░░░░░░
Articles done:0 / 70
Complete:0%
Notes written:0
Highlights:0