This experiment should answer the following key question:
Can an agent learn to explore its environment from curiosity alone?
In this question are a couple of sub-questions:
How much processing power is required?
How long does it take?
What metric can we use to quantify this?
This is a difficult question to answer because I am not currently aware of a way to quantify "good" exploration.
In my mind, if we can figure out a way to quantify the diversity of experience the agent encounters then this should be sufficient to answer the proposed question.
This experiment should answer the following key question:
In this question are a couple of sub-questions:
This is a difficult question to answer because I am not currently aware of a way to quantify "good" exploration.
In my mind, if we can figure out a way to quantify the diversity of experience the agent encounters then this should be sufficient to answer the proposed question.