Abstract:
The predictive processing theory states that generative models make predictions on future inputs, and those models are assumed to get increasingly complex in developing infants, but how this occurs is not yet fully understood. With the robo-havioral methodology, the behavior of the k-means method to define the hypothesis space are inspected and experience driven parameter is used to refine this space. No advantages were found in the clustering with k-means, while fine-graining of certain areas in the hypothesis space using the accumulation of experience helped the goal-oriented robot to further decrease its error.
