Resilient aggregation with attack detection in sensor networks

In this paper, we propose a new model of resilient data aggregation in sensor networks, where the aggregator analyzes the received sensor readings and tries to detect unexpected deviations before the aggregation function is called. In this model, the adversary does not only want to cause maximal distortion in the output of the aggregation function, but it also wants to remain undetected. The advantage of this approach is that in order to remain undetected, the adversary cannot distort the output arbitrarily, but rather the distortion is usually upper bounded, even for aggregation functions that were considered to be insecure earlier (e.g., the average). We illustrate this through an example in this paper