Analogue winner-take-all neural networks for determining maximum and minimum signals

The ability to identify the signal with the maximum or minimum instantaneous value is central in many real-time applications. Two types of winner-take-all neural networks for determining maximum and minimum signals on-line and in parallel are presented. Starting with a review of the existing winner-take-all neural networks, the paper proposes a type of laterally-inhibited neural network without self state feedbacks and a type of comparator-based neural network with self state feedbacks. Each competitive winner-take-all network consists of O(n) neurons and O(n²) connections, and each comparator-based winner-take-all network consists of O(n) neurons and connections. Applications of the proposed winner-take-all networks to sorting are also discussed. © 1994 Taylor & Francis Ltd.