1: \begin{abstract}

2: This paper develops a mathematical and computational framework for

3: analyzing the expected performance of Bayesian data fusion, or joint

4: statistical inference, within a sensor network. We use variational

5: techniques to obtain the posterior expectation as the optimal fusion

6: rule under a deterministic constraint and a quadratic cost, and study

7: the smoothness and other properties of its classification performance.

8: For a certain class of fusion problems, we prove that this fusion

9: rule is also optimal in a much wider sense and satisfies strong asymptotic

10: convergence results. We show how these results apply to a variety

11: of examples with Gaussian, exponential and other statistics, and discuss

12: computational methods for determining the fusion system's performance

13: in more general, large-scale problems. These results are motivated

14: by studying the performance of fusing multi-modal radar and acoustic

15: sensors for detecting explosive substances, but have broad applicability

16: to other Bayesian decision problems.

17: \end{abstract}

18: