Dr. Anna Vaskevich and Prof. Liz Torres report in Frontiers in Neuroscience a neural correlate of a new form of random exploratory statistical learning common to neonates, but overlooked by current models of learning. Whereas traditional learning relies on a reference signal always provided to the system, to optimize prediction error, and minimize surprise, this new mode of learning, ascribes equal probability to all future events and operates under high signal regime, open to surprise. In a novel environment, the brain can thus autonomously self-discover the reference signal that it needs to enter into the type of error correction based learning mode, amenable to build priors. This work explains how a naïve system can autonomously learn.
Read the full article in Frontiers in Neuroscience: Rethinking statistical learning as a continuous dynamic stochastic process, from the motor systems perspective