How to Combine Prior Knowledge With New Evidence

Bayes’ rule

One of the most common tasks we will encounter in the following lesson is the integration of prior knowledge with new evidence. For example, we might have a robot that estimates its current position at certain values and then gets new (noisy) sensory data that adds some suggestions for different positions. This is also a common task in the fusion of data from different sensors. In general, we assume that we already have a model that we built from previous data, and now we want to refine this model with new data. The general question we have to solve is how to weigh the different evidence in light of the reliability of this information. Solving this problem is easy in a probabilistic framework and is one of the main reasons that so much progress has been made with the application of probabilistic machine learning.