Multi-Armed Bandit: Explore vs Exploit
N=1000 tokens. 5 arms with unknown payouts. Best strategy minimizes regret.
Strategy:
ε-first (explore then exploit)
ε-greedy (mixed)
UCB (Upper Confidence Bound)
ε =
0.10
Run 100
Run All 1000
Reset
Reveal True Rates
100 Trials (MC)
Tokens Used
0
Total Reward
0
Best Possible
0
Regret
0
Current Phase
—
Cumulative Reward
Cumulative Regret
Regret Distribution (100 independent MC trials)
Click "100 Trials (MC)" to run Monte Carlo simulation.