Generation, recognition and learning of recurrent signals by pulse propagation networks

Pulse propagation networks (PPN) are neural networks in which individual ac tion potentials encode information. The dynamics of PPN depend not only on the synaptic weights of connections but also the delay in the propagation o f action potentials between neural elements. It is known that PPN can perfo rm complex computations and information processing by encoding information as the time intervals between action potential events. In this paper we app roach the practical question of constructing PPN to generate, recognize and learn arbitrary recurrent signals. We present specific examples of network s that generate and recognize signals and also describe a learning algorith m that allows PPN to learn by self-organization. Finally we discuss the pos sible importance of dynamical fluctuations about the mean-activity field of a neural network.