Visualizing the Loss Landscape of Actor Critic Methods with Applications in Inventory Optimization

Continuous control is a widely applicable area of reinforcement learning. The main players this are actor-critic methods that utilize policy gradients neural approximators as common practice. focus our study to show the characteristics actor loss function which essential part optimization. We exploit low dimensional visualizations and provide comparisons for landscapes various algorithms. Furthermore, we apply approach multi-store dynamic inventory control, notoriously difficult problem in supply chain operations, explore shape associated with optimal policy. modelled solved using learning while having landscape favor optimality.