Common use of Experiments Clause in Contracts

Experiments. We investigated whether the correct diagnosis is among the most probable diagnoses if the knowledge of one of the agents contains an error. In the experiments, we gen- erated 8000 systems each to be diagnosed by three agents. We chose three agents since this is the smallest number to make one diagnosis significantly more probable if one of the agents disagree with the others, while using more agents would have simplified the diagnostic problem. Each gener- ated system consisted of 40 components, each with one out- put and two inputs. An input was either connected to one of the four system inputs or to an output of a randomly cho- sen component without causing cycles. The normal behavior of a component was a modulo n adder for each of the three agents each using a different perspective. Besides, a component had faulty behaviors, namely ab and two specific faulty behaviors f1 and f2. In both fault modes f1 and f2, a fault value was added mod- ulo n to the output of the component. These faults values were randomly chosen for each combination of a compo- nent, a fault mode and an agent. Finally, for every compo- nent c, the same value was used for the probabilities of the fault modes f1 and f2 and the probability that a behavior mode is incorrect. To create a diagnostic problem, in each generated sys- tem one component was chosen to be the broken compo- nent and one of the fault mode f1 or f2 was selected for the component. In one of the three perspectives, however, the component behaved according to the other fault mode, i.e. the knowledge of the agent using this perspective was in- correct in the current situation.