Design and Analysis of Many-to-Many Associative Memory Driven by External Input

The many-to-many associative memory is a brain-inspired functionality, which includes one-to-many and many-to-one associations. Most models neglected the inherent relationships of the input patterns or output patterns. To address this issue, this paper proposes a cascaded multivalued continuous-time recurrent neural networks, where the current network's output is linked with both its own input and the output of the previous network. The global asymptotic convergence of equilibrium points is leveraged to design a learning algorithm based on inequalities. Compared to multidirectional associative memory networks, this model can reflect the temporal and inherent relationships of the inputs or outputs, requires fewer neurons for the associative process, and imposes no constraints on the initial state of the neurons.