Usefulness of LSTM for Reward Prediction (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
To eliminate the dependence of the reward function on the student state, Long Short-Term Memory (LSTM) networks were used to predict the probability of a student answering correctly. Surprisingly, on generated data, the SuperMemo algorithm performed significantly better than an LSTM trained on interaction data generated by two other models. This finding supports the idea that higher-quality interaction data leads to better LSTM performance. Additionally, LSTMs demonstrated better performance than the average sum of outcomes reward function.
0
1
Contributors are:
Who are from:
Tags
Data Science
Related
Summary of findings (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
How useful is average of sum of outcomes based reward function? (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Limitations (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Sustainability and Ethics (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Future Work (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)
Usefulness of LSTM for Reward Prediction (Using deep reinforcement learning for personalizing review sessions on e-learning platforms with spaced repetition)