Reinforcement Learning and Artificial Intelligence (RLAI)
Topics that Rich sees as excellent research opportunities and essential to solving AI
Non-Markov modeling (need to construct approximations to state)
PSR's, OOMs, TD nets
Knowledge representation, including
temporal abstraction (options/TD nets)
predictive representations and TD nets again
discovery (of useful predictions)
Using the model - planning
sample-based methods
Off policy learning with bootstrapping and linear function approximation
Discovery (options, rep'ns, features, states, questions...)
"Real" applications (in the sense of Michael Littman)
Scale up (but could still be a toy problem - sorry Michael)
Curiosity -- Intrinsic/changing reward, incl exploration/exploitation
Dynamic prioritization (meta control)
intentionality
multiple drives, motivational states
Generalization, similarity
predictive representations again
