Multi-Armed Bandit
A classic framework in sequential decision making in which the decision maker selects one of several arms to pull in each round and receives a random reward. It requires balancing exploration and exploitation to maximize cumulative reward.
Timeline (last 90 days)
In a shared budget allocation scenario, if there are K segments and each segment's weight is chosen from M discrete values, the total number of configurations is M^K, leading to a combinatorial action space explosion.
Classic multi-armed bandit algorithms (such as ε-greedy) assume that the reward distribution of each arm is independent; when the action space explodes combinatorially and feedback is sparse, the exploration cost of independently evaluating each arm grows linearly with the number of actions, and efficiency drops sharply.
All Facts (2)
In a shared budget allocation scenario, if there are K segments and each segment's weight is chosen from M discrete values, the total number of configurations is M^K, leading to a combinatorial action space explosion.
50%UnverifiedClassic multi-armed bandit algorithms (such as ε-greedy) assume that the reward distribution of each arm is independent; when the action space explodes combinatorially and feedback is sparse, the exploration cost of independently evaluating each arm grows linearly with the number of actions, and efficiency drops sharply.
50%