Dynamic Cache Placement and Trajectory Design for UAV-Assisted Networks: A Two-Timescale Deep Reinforcement Learning Approach