Experimental design for evaluating TRPO performance with reward shaping (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
To evaluate TRPO's performance with reward shaping, the number of episodes per run was fixed at 40 (all other experiment parameters unchanged) using the EFC environment. The LSTMs supplying the reward signal were trained separately on three data sources to compare how training-data quality affected agent performance: (1) a random sample, (2) a random-policy tutor, and (3) a SuperMemo tutor.
0
1
Contributors are:
Who are from:
Tags
Data Science
Related
Experimental Setup (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Training the LSTM (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Relation between rewards and thresholds (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Performance of RL agent when the number of items are varied (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Performance of TRPO vs. TNPG algorithms (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Comparison between likelihood and average of sum of outcomes based reward functions (research objective) (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Evaluation (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Reward functions and performance metrics (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Experimental design for evaluating TRPO performance with reward shaping (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)