Pavlov and Hume Learning Applied to a Neural Network: A neural simulation study within Connectology

Abstract

The neocortex has an amazing ability to associate constant changing sensory impressions from different sources. It is assumed that the association property is acquired when the synapse strength between the neocortex's neurons are altered. In Connectology, a connectionist theory of brain-psychology, it is purposed three learning mechanisms for synaptic alteration. The mechanisms, Skinner learning, Pavlov learning, and Hume learning are based on the respective principles; Hedonism - learning from award, Anticipation - human reasoning, and Reason - object recognition. This Master's project addresses whether the combination of Pavlov and Hume learning is able to explain the ability to learn to associate different sense impressions in neocortex. The learning mechanisms will be applied to an ac{ANN} to explore whether they are able to recreate parts of the neocortex's association ability, by associating two changing sensory impressions, represented as 2D-patterns. The findings show that the Pavlov and Hume learning need guidance to be able to associate the different changing sensory impressions. Making the ac{ANN} small enough to communicate directly between the changing 2D-patterns, will function as a type of guidance, and the ac{ANN} is able to associate up to about 50 patterns with only the use of the Pavlov and Hume learning. Thus cannot Pavlov and Hume learning on their own explain the neocortex's ability to learn to associate different sensory impressions, but can help understand parts of this capability.