A proof of convergence of inverse reinforcement learning for multi-objective optimization
This provides theoretical guarantees for IRL in multi-objective settings, which is incremental as it extends existing convergence proofs to new methods and formulations.
The paper tackled the convergence of inverse reinforcement learning (IRL) methods for multi-objective optimization by proving convergence for Wasserstein IRL using the projective subgradient method and for other IRL methods like maximum entropy IRL with gradient descent.
We show the convergence of Wasserstein inverse reinforcement learning for multi-objective optimizations with the projective subgradient method by formulating an inverse problem of the multi-objective optimization problem. In addition, we prove convergence of inverse reinforcement learning (maximum entropy inverse reinforcement learning, guided cost learning) with gradient descent and the projective subgradient method.