Home▸AI & Machine Learning▸Multi-Armed Bandit — Marketing Budget Allocator

Multi-Armed Bandit — Marketing Budget Allocator

Interactive multi-armed bandit simulator: watch epsilon-greedy and UCB1 algorithms allocate a marketing budget across five ad channels in real time, learning which channel converts best while paying out cumulative regret.

AI & Machine Learning3DModerate60 FPS📱 Mobile-adapted⇄ 2D version

This simulator turns a classic reinforcement-learning problem — the multi-armed bandit — into a live marketing-budget dashboard. Five ad channels each hide a true, fixed conversion rate the algorithm cannot see. Every simulated budget unit is a "pull": a token flies from the budget source to the channel the current algorithm chose, and lands as a win (green) or a miss (red) sampled from that channel's hidden rate. Switch between epsilon-greedy exploration, UCB1's confidence-bound exploration, and a random baseline, and watch the bar heights — each channel's running estimate of its own conversion rate — converge toward the truth while cumulative regret tracks how much conversion value was spent finding the winner.

⚙ Under the hood

Watch epsilon-greedy, UCB1 and a random baseline compete to allocate a marketing budget across five ad channels with hidden conversion rates, tracking live spend, conversions and cumulative regret in 3D.

machine learningreinforcement learningmulti-armed banditbusiness decisionsUCB1marketing analytics

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install