On noiseless diagnosis

Let F be the fault of a system which takes value in /spl Omega/={f/sub i/}, and p/sub i/ be the known probability of occurrence of f/sub i/. Let T={t/sub j/} be a sufficient set of tests available for diagnosing the system, and c/sub j/ be the cost of t/sub j/. The number of possible responses for each test in T may be different. The author introduces the cost-entropy function as an information-theoretic lower bound on C/sub min/ the expected cost of an optimal testing tree. The author also obtains a universal upper bound of C/sub min/ when {p/sub i/} is unknown, and a refined upper bound on C/sub min/ when {p/sub i/} is known. The author's results are essential for developing heuristic strategies to search for optimal and suboptimal testing trees.<
>