Meta-level operations Polus Framework Problem determination

Base-level reasoning

Input: The set of queries $\phi$derived by problem determination
Output: A set of candidate_actions that partially or completely satisfy elements in $\phi$.
Approach:
The logic for searching the knowledge base is expressed in first-order predicate calculus. The logic captures the thought process that is implicit while writing imperative specifications. The information model of the action object makes it possible to express these semantics and derive the actions to be invoked ``on-the-fly.'' A few examples of the thought process, expressed as first-order predicates, are described below. At the high-level, a candidate action is one that affects the component in $\phi$, satisfies the preconditions ($p$) in the current-state, and has the desired implications ($i$).

$$\:\{\:a \:\epsilon \:Action,\; s \:\epsilon\: State\}$$

$$\begin{gather*}\begin{split} & \forall \: a,s,\phi\;candidate (a,s,\phi)\: \Rig... ...edge \; \neg (Exclusion\_set(y) \:\epsilon\: Value(s)) \end{split}\end{gather*}$$

At each step in the reasoning process, while selecting the candidate-actions, Polus maintains a log of the available choices and the option that was selected. This can later serve as an explanation to the human administrator, regarding how a specific action was selected. In the example, assume the two candidate actions that are selected based on Query 1 and 2 are: 1. Invoking prefetching at the client machine 2. Invoking data replication at the disks

Meta-level operations Polus Framework Problem determination