Mapping binary associative memories onto sigmoidal neural networks using a modified projection learning rule

This paper shows the applicability of the well-known projection learning rule to the design of associative memories based on continuous-time neural networks, with sigmoidal nonlinearities. The proposed design method exhibits several interesting features: learning capability, computational efficiency, exact storage of binary vectors as asymptotically stable equilibrium points, and global stability of the resulting network. An example is included to illustrate the method