Beyond State-Wise Mirror Descent: Offline Policy Optimization with Parametric Policies

1. 1Investigate theoretical aspects of offline reinforcement learning under general function approximation
2. 2Develop algorithms that are computationally tractable for offline policy optimization
3. 3Apply pessimism to learn a good policy from offline data
4. 4Extend existing algorithms, such as PSPI, to handle large action spaces

This research benefits AI engineers and ML researchers working on reinforcement learning, as it provides a foundation for offline policy optimization with general function approximation, which can be applied to complex decision-making problems

💡 Offline reinforcement learning can be achieved with general function approximation, enabling more complex decision-making problems to be solved

💡 Offline RL policy optimization with parametric policies