Model Selection for Production System via Automated Online Experiments
This addresses a practical challenge for machine learning practitioners in industry by enabling efficient model selection with limited online experiments, though it is incremental as it builds on existing online experimentation and Bayesian methods.
The paper tackles the problem of selecting the best model for deployment in production systems, where traditional A/B tests are limited by budget constraints, by proposing an automated online experimentation mechanism that efficiently identifies the optimal model from a large pool using a Bayesian surrogate model and exploration-exploitation strategies, demonstrating effectiveness in simulations based on real data.
A challenge that machine learning practitioners in the industry face is the task of selecting the best model to deploy in production. As a model is often an intermediate component of a production system, online controlled experiments such as A/B tests yield the most reliable estimation of the effectiveness of the whole system, but can only compare two or a few models due to budget constraints. We propose an automated online experimentation mechanism that can efficiently perform model selection from a large pool of models with a small number of online experiments. We derive the probability distribution of the metric of interest that contains the model uncertainty from our Bayesian surrogate model trained using historical logs. Our method efficiently identifies the best model by sequentially selecting and deploying a list of models from the candidate set that balance exploration-exploitation. Using simulations based on real data, we demonstrate the effectiveness of our method on two different tasks.