Emergence of Sensory Representations Using Prediction in Partially Observable Environments

n order to explore and act autonomously in an environment,an agent can learn from the sensorimotor information that is capturedwhile acting. By extracting the regularities in this sensorimotor stream,it can build a model of the world, which in turn can be used as a basis foraction and exploration. It requires the acquisition of compact representa-tions from possibly high dimensional raw observations. In this paper, wepropose a model which integrates sensorimotor information over time,and project it in a sensory representation. It is trained by preformingsensorimotor prediction. We emphasize on a simple example the role ofmotor and memory for learning sensory representations