MABs

Multi-Arm Bandits (MABs) with i.i.d, non-stationary, and federated bandit examples.

The paper (Paper_on_Projects_1_and_2.pdf) describes the first two parts of the experiment.

The presentation (Federated MAB.pptx) describes the federated bandit experiment.

Projects:

1. 10-armed Testbed (I.I.D.)
2. Applying MABs to non-I.I.D. ad optimization simulation.
3. Built a federated MAB framework based off of the Federated Averaging algorithm proposed in Communication-Efficient Learning of Deep Networks from Decentralized Data (McMahan et al.). This was then applied to the ad optimization simulation, where the server should not know the exact clicks of its users.