Next: Category 5: Negation of
Up: Experimental Analysis
Previous: Category 3: More than
|
Figure: Comparing Polus and ECA for category 4 (recurrent action invocation). In the graph, the latency goal = 4.5 msec |
Analysis: The comparison in shown in figure 12. Polus uses look ahead while invoking actions to estimate the impact of the action on the goals. Rule-based systems don't have an equivalent of this (though it is possible to write separate rules to cater for this). As shown in the graph, ECA invokes the Backup action that leads to invocation of a series of actions (Replication in this example). Polus does a look-ahead and does not invoke Back-up during the current system-state. ( For Back-up we are assuming a time-window) Insights: Look-ahead is a required operation and effective only with some learning of the action model. Hence there is only a single curve in the graph.
