Aidan Gleich's Adaptive Policy Learning Tackles Unknown Network Interference
New algorithm optimizes treatment allocation and learns interference dynamics simultaneously.
Aidan Gleich, Eric Laber, and Alexander Volfovsky introduce 'Adaptive Policy Learning Under Unknown Network Interference'. Their Thompson sampling algorithm jointly learns the interference network and optimizes individual-level treatment allocations via a Gibbs sampler. The method returns both an optimized treatment policy and an estimate of the interference network, supporting downstream causal analyses such as estimation of direct, indirect, and total treatment effects. Empirically, it achieves more than an order-of-magnitude reduction in regret. On two real-world networks, the algorithm achieves sublinear regret and yields downstream effect estimates with small RMSE relative to the truth.
- Developed a Thompson sampling algorithm for optimizing treatment allocations under interference.
- Achieved over an order-of-magnitude reduction in regret in empirical tests.
- Supports accurate estimation of treatment effects for better decision-making.
Why It Matters
This advancement empowers professionals to make data-driven decisions in complex environments.