If you don't want to explicitly create a reward function, what options do I have?

General
1

If you don't want to explicitly create a reward function, what options do I have? (that are compatible with your courses in Deep RL*)

2

Hi Emma,



You can explore methods like Imitation Learning, where agents mimic certain sources by learning from historical data, as demonstrated in the Imitative Market Maker (IMM) framework. Here is a research paper for your reference. Another method is Inverse Reinforcement Learning (IRL), which infers the underlying reward function by observing expert behavior, allowing agents to replicate successful strategies. You can go through this paper to get more insights for the same.





Hope this helps!