foundations of computational agents
The Q-learner controller does not specify what the agent should actually do. The agent learns a -function that can be used to determine an optimal action. There are two things that are useful for the agent to do:
exploit the knowledge that it has found for the current state by doing one of the actions that maximizes
explore in order to build a better estimate of the optimal -function; it should sometimes select a different action from the one that it currently thinks is best.
There have been a number of suggested ways to trade off exploration and exploitation:
The -greedy exploration strategy, where , is the explore probability, is to select the greedy action (one that maximizes ) all but of the time and to select a random action of the time. It is possible to change through time. Intuitively, early in the life of the agent, it should act more randomly to encourage initial exploration and, as time progresses, it should act more greedily (reducing ).
One problem with an -greedy strategy is that it treats all of the actions, apart from the best action, equivalently. If there are a few seemingly good actions and other actions that look much less promising, it may be more sensible to select among the good actions: putting more effort toward determining which of the promising actions is best, and less effort to explore the actions that look worse. One way to do that is to select action with a probability depending on the value of . This is known as a soft-max action selection. A common method is to use a Gibbs or Boltzmann distribution, where the probability of selecting action in state is proportional to . Thus in state , the agent selects action with probability
where is the temperature specifying how randomly values should be chosen. When is high, the actions are chosen in almost equal amounts. As the temperature is reduced, the higher-valued actions are more likely to be chosen and, in the limit as , the best action is always chosen.
An alternative is optimism in the face of uncertainty: initialize the -function to values that encourage exploration. If the -values are initialized to high values, the unexplored areas will look good, so that a greedy search will tend to explore. This does encourage exploration; however, the agent can hallucinate that some state–action pairs are good for a long time, even though there is no real evidence for them being good. A state only gets to look bad when all its actions look bad; but when all of these actions lead to states that look good, it takes a long time to get a realistic view of the actual values. This is a case where old estimates of the -values can be quite bad estimates of the actual -value, and these can remain bad estimates for a long time. To get fast convergence, the initial values should be as close as possible to the final values; trying to make them an overestimate will make convergence slower. In noisy environments, optimism in the face of uncertainty with no other mechanism for exploration, can mean that a good action never gets explored more because, by random chance, it gets a low Q-value from which it cannot recover.
It is interesting to compare the interaction of the exploration strategies with different choices for how is updated. See Exercise 3.