Solving Complex Control Tasks via Simple Rule(s): Using Chaotic Dynamics in a Recurrent Neural Network Model

Novelty-induced memory transmission between two nonequilibrium neural networks

“…The problem resides in how to describe these dynamics, which are associated with the proceeding of adaptive learning. Remarkably, various experiments and theories have suggested that chaos is crucial for learning (Tsuda et al 1987;Skarda and Freeman 1987;Tsuda 1991Tsuda , 1992Nara and Davis 1992;Sano 2000;Tsuda 2001;Kay 2003;Kozma and Freeman 2001;Raffone and van Leeuwen 2003), and recent studies on autonomous robots have shown that chaotic neural dynamics could be potentially useful to solve complex ill-posed problems via simple rule(s) (Li et al 2007;Li and Nara 2008a, b;Yoshida et al 2010;Li and Nara 2012). These studies implied that chaotic neural dynamics could play an important role in adaptively coping with the onset of uncertainty from the ever-changing environment.…”