No abstract

No abstract

Cognition is a fundamental feature of natural intelligence, which a modern technology has not yet been able to reproduce in full capacity. Sensor networks provide a new technological support for a substantial increase in an amount and quality of information that might be collected and communicated in complex adaptive systems. Their application may significantly raise the degree of intelligence in system design and implementation into the levels where effects of cognition will start kicking in. The paper describes the results of an empirical study aiming to demonstrate that a cognition ability may be treated as a generic sensor network feature. The new architecture with neural networks distributed over the sensor network platforms was developed for sensor network engineering applications. The detection system learns to detect the change of not only the signal levels but also sensor signal shapes and parameters that represent a more complicated task. The architecture allows for a significant reduction in resource consumption without compromising the change detection performance. Implemented as an agent controlling the sensor network self-adjustment to the objects under measurement in the sensor network composed from typical sensor motes, the novel neural network structures may achieve a significant saving in power consumption and an increase in a possible network deployment time from a few days to a few years. The experiments prove that a neural-network-based change detection system is feasible for sensor networks application designs and could be successfully implemented on the technological platforms currently available on the market.Index Terms-Artificial neural networks (ANNs), distributed artificial intelligence, sensor networks, signal change detection.

No abstract

This paper proposes a novelty detection system for sensor networks. The system utilizes a neural network function prediction methodology to predict sensor outputs in order to determine if the sensor outputs are novel. In addition to the novelty detection system, a modification to a standard neural network function predictor is proposed that allows the novelty detection system to quickly learn to accurately predict next sensor outputs. The parameter choice and the relationship between the threshold values and false alarm and missing attack rates are studied and recommendations are provided.