Figure 4
From: Robustness of quantum reinforcement learning under hardware errors
Comparison of Q-learning with shot noise using the informed shot-allocation method (labeled “max shots”) proposed in this work, and a standard measurement scheme that simply assigns a fixed number of shots to each circuit evaluation (labeled “shots”). Results are averaged over ten agents for each configuration. (a) Shows results for agents trained in CartPole environment, (b) shows results for agents in the TSP environment