Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation

1. 1Implement transformer policies trained on synthetic data to estimate arm-conditional outcome variances
2. 2Use the estimated variances to inform covariate-dependent Neyman rule allocations
3. 3Evaluate the performance of the Bayesian in-context experimenter using simulated experiments
4. 4Compare the results to traditional oracle design methods
5. 5Refine the transformer policy using in-context learning to improve estimation accuracy

Data scientists and AI engineers can benefit from this approach to improve the efficiency of their experiments, while researchers can use it to develop more accurate models

💡 Transformers can amortize the sequential variance-estimation and allocation process in adaptive experiments, improving statistical efficiency

🤖 Transformers can be used as Bayesian in-context experimenters for efficient ATE estimation! 💡