The theme of this year's workshop is Representation Sculpting for Life-Long Learning. The idea is that a core competence for a long-lived reinforcement learning agent is the ability to gradually construct and shape powerful representations for future learning, generalization, and planning. This theme includes topics such as:

perceptual learning
unsupervised learning from rich temporal data streams
construction of state representations
feature selection and construction for linear function approximation
discovery of useful options and option models (ways of behaving and beliefs about their outcomes)
feature de-correlation
roles of sparsity in feature representations
information theoretic approaches to feature construction
Kanerva coding
cascade correlation networks
online cross validation
intrinsic motivation and computational curiosity
feature representations specifically designed to support predictions