Multi-variate Distributed Data Fusion with Expensive Sensor Data

Distributed fusion of complex information is critical to the success of large organizations. For such organizations, comprised of thousands of agents, improving and shaping the quality of conclusions reached is a challenging problem. The challenge is increased by the fact that acquisition of information could be costly. This leads to the crucial requirement that the organization should strive to reach correct conclusions while minimizing information acquisition cost. In this paper, we have developed a model of complex, interdependent information that is costly to acquire and where complex fusion should be optimized within an organization while minimizing the cost of acquiring the sensor data. Empirical results show a number of interesting effects. First, unselfish agents who spend resources (even when not strictly locally necessary) can lead to substantial improvement in the overall accuracy of the organization's conclusions. Second, an organization can substantially improve its performance by carefully assigning sensor resources within the organization. Third, over time, agents can learn the reliability of the members of the organization to whom they are directly connected to improve performance. Learning can also lead to better team decisions about whether to spend resources and how much resource to expend to get sensor data. Our conclusions and algorithms can help a range of organizations reach better conclusions while expending less resources procuring sensor data.