Offline Reinforcement Learning

2 selectedDifficulty 7-82 unseenView topic

AdvancedNew

0 answered

2 advancedAdapts to your performance

Question 1 of 2

120sadvanced (7/10)conceptual

In offline RL on a fixed dataset D, why does naive Q-learning often catastrophically overestimate Q (s, a) for out-of-distribution actions a ?