Simultaneous verses successive learning in neural networks

Abstract

Neural networks were inspired by the human brain, with great hopes that neural networks would capture the vast potential of its biological counterpart. This paper explores the link between neural networks and the human brain in the context of simultaneous vs. successive learning. Learning experiments conducted on human subjects were modeled and repeated using neural networks as test subjects. Neural networks confirmed the conclusion from human subject experiments that simultaneous learning was faster than successive learning. Loess and Duncan further extended their hypothesis without formal experimental evidence that simultaneous would outperform successive as complexity increased beyond the scope of their human experiments. Interestingly, neural networks contradict their hypothesis. The results from neural networks demonstrate an existence of a threshold, after which the effects of simultaneous and successive learning become negligible. Intuitively, when humans are presented with complicated tasks, the type of learning is immaterial, since the complexity of the problem would overwhelm any advantages one method has over the other. Confirmation of this intuition can only be confirmed through future human experiments. Furthermore, this paper demonstrates that neural networks can be used as a rough model and give valuable insight into a problem, before the costly human subject experiments are conducted.