In this paper we present a theoretical framework for novelty based feedback regulation in artificial neural networks. Novelty is assessed on the basis of monitoring the coherence of network dynamics. The result of novelty detection is dynamically coupled to parameters that control the dynamics of the recognition process. The paper presents a new measure of novelty detection - the strength of the local field - and presents new simulation results concerning novelty detection. It also integrates previously published models and simulation results into a general dynamical model of feedback regulation.