Bifurcation analysis on a generalized recurrent neural network with two interconnected three-neuron components (Q2393214)

A class of recurrent neural networks is constructed with two interconnected \(n\)-neuron components, which generalizes a specific class of \(n\)-neuron networks introduced by \textit{A. C. Ruiz}, \textit{D. H. Owens} and \textit{S. Townley} [``Existence, learning and replication of periodic motions in recurrent neural networks'', IEEE Trans Neural Netw. 9, 651--661 (1998)]. The authors consider the case of \(n=3\), for which the network can be described by the following system of ordinary differential equations: \[ \begin{aligned} &\dot{x}_1(t)=-x_1(t)+\tanh(x_2(t)),\\ &\dot{x}_2(t)=-x_2(t)+\tanh(y_3(t)),\\ &\dot{x}_3(t)=-x_3(t)+w_1\tanh(x_1(t))+w_2\tanh(x_2(t)),\\ &\dot{y}_1(t)=-y_1(t)+\tanh(y_2(t)),\\ &\dot{y}_2(t)=-y_2(t)+\tanh(x_3(t)),\\ &\dot{y}_3(t)=-y_3(t)+v_1\tanh(y_1(t))+v_2\tanh(y_2(t)), \end{aligned}\tag{1} \] where \(x_i(t)\in\mathbb{R}\) and \(y_j(t)\in\mathbb{R}\), \(i,j=1,2,3\) are the state variables, and \(w_i\in\mathbb{R}\), \(v_i\in \mathbb{R}\), \(i,j=1,2\) are the network weights. It is shown that system (1) experiences generic pitchfork and Hopf codimension one bifurcations. It is also proved that the emergence of generic Bogdanov-Takens, pitchfork-Hopf and Hopf-Hopf codimension two, and the degenerate Bogdanov-Takens bifurcation points in the parameter space is possible due to the intersections of codimension one bifurcation curves. The occurrence of bifurcations of higher codimensions significantly increases the capability of the newly constructed recurrent neural network to learn broader families of periodic signals.