next up previous contents
Next: Probabilistic and Bayesian Machine Up: Discussion and Properties Previous: Perception and Expression

Imitation Learning

In cognitive science work, imitation style learning has been investigated as a source of higher order intelligence and fast acquisition of knowledge [50]. The ARL model is also intimately coupled with the notions of imitation learning. There is no exhaustive search in the acquisition of behaviour but rather a direct relationship between teachers and student through the interactions that were demonstrated. Here, the teachers are humans interacting with each other and the student is the ARL system. There is no truly laborious teaching on the behalf of the humans since they merely interact naturally. In essence, the learning that is taking place involves no real programming or data processing effort for the humans. There is no manual classification, labeling, segmentation, reinforcement or structure identification in this scenario. Observations are acquired and learned autonomously by the system. The model is fully estimated using standard, computationally tractable probabilistic approaches and exploits tried and true Bayesian statistical formalisms.

Tony Jebara