Hyperparameter Selection for Imitation Learning
This addresses a practical issue for researchers and practitioners in imitation learning, as it provides a solution for hyperparameter selection in realistic settings where reward functions are not observable, though it is incremental in focusing on a specific bottleneck.
The paper tackles the problem of tuning hyperparameters for imitation learning algorithms when the expert's reward function is unavailable, proposing and evaluating proxies for reward in an empirical study with over 10,000 agents across 9 environments, showing that good hyperparameters can often be selected using these proxies.
We address the issue of tuning hyperparameters (HPs) for imitation learning algorithms in the context of continuous-control, when the underlying reward function of the demonstrating expert cannot be observed at any time. The vast literature in imitation learning mostly considers this reward function to be available for HP selection, but this is not a realistic setting. Indeed, would this reward function be available, it could then directly be used for policy training and imitation would not be necessary. To tackle this mostly ignored problem, we propose a number of possible proxies to the external reward. We evaluate them in an extensive empirical study (more than 10'000 agents across 9 environments) and make practical recommendations for selecting HPs. Our results show that while imitation learning algorithms are sensitive to HP choices, it is often possible to select good enough HPs through a proxy to the reward function.