Trading off the Costs of Inference vs. Probing in Diagnosis

Th is paper proposes a new a lgor i thm which when provided the relative costs of computa­ t ion vs. probing minimizes the to ta l cost of diagnosis. Dur ing the diagnosis process the de­ cision of whether to probe or to compute is dependent on the expected costs and benefits of each al ternat ive. I t is unl ikely tha t we w i l l be able to find general analyt ic and simpleto-compute models for the costs and benefits. Therefore, we base our a lgor i thm on simple em­ pir ical ly derived models of costs and benefits. W i t h these models, our a lgo r i thm operates by continuously choosing the o p t i m u m action to make next. Th is a lgo r i thm w i l l not blow up on the rare pathological cases and w i l l always (on average) find diagnoses at equal to or better cost than a conventional GDE/Sher lock . When the cost of probing is h igh , then our a lgor i thm behaves exactly the same as GDE/Sher lock . When the cost of computa t ion is h igh, the algo­ r i t h m performs the diagnosis at far lower cost than GDE/Sher lock .