Results (
Thai) 1:
[Copy]Copied!
It is important to understand the distinction between a decisionmade with low con fidence andadecision made basedon low quality information. Ideally, we would like these to bethe same: when the information is poor, the decision shouldbe relatively ambiguous. In particular, when the informationfrom one modality is poor, that modality should have relativelylittle inuence on the overall decision. However, in many cases, just the opposite is true. In the case describedabove | a model poorly trained for the environment in whichit is operating | low probability estimates tend to be accompaniedby inappropriately high con dence. The reasonfor this is that probabilities of independent observations are multiplied together. A single outlier in the observation sequence,which has a very low probability in all states, canexert more inuence on the nal decision than the high probabilityobservations.
Being translated, please wait..