This paper covers a system for sensor interpretation. Interpretation involves the determination of high-level explanations of observations. Not sure about the relevance of the paper other than to show the complexity of using sensor data.

Expansion of the Blackboard technique, construct alternative hypothesis and then prove or disprove it.

New strategies: differential diagnosis.

2 key components:

1. evidential representation that includes explicit symbolic encodings of the sources of uncertainty in the evidence for hypotheses.

2. a script based, incremental control planner.

A distinction is drawn between *Classification* and *Constructive* problem solving. In classification, the interpretation is chosen from a
pre-enumerated set of possible solutions.
In constructive, the set of possible solutions is determined as part of
the problem solving process.

Authors note:

Classification is good for diagnosis.

Constructive is good for interpretation.

Two types of interpretation processes:

1. classification problem solving: selection from pre-known alternatives. Techniques such as Dempster-Shafer and Bayesian networks work well in this context.

2. constructive problem solving: solutions determined in the solving of the problem. Blackboard model works well for this because it supports opportunistic control for dealing with uncertain data and problem solving knowledge. However, in practice it is limited to hypothesize and test strategies.

blackboard- most implementations limited to incremental hypothesis and test strategies.

Things to know:

blackboard framework

differential diagnosis

evidential representation

abductive reasoning

Dempster-Shafer

Bayesian network

hypothesize and test strategies.