Flexible strategy coordination approach for supporting resource awareness in multi agent filtering framework

Speeding Coordination by Combining Analytical and Inductive Learning

“…The selection of such a strategy will produce long lasting filtering job [13], which will usually find perfect results, but unfortunately after the unacceptable delay. Since it is hardly possible that anyone will be patient enough to wait for them, it is much better to offer nearly perfect results, obtained after a reasonable duration of filtering.…”

“…Since it is hardly possible that anyone will be patient enough to wait for them, it is much better to offer nearly perfect results, obtained after a reasonable duration of filtering. Taking care of the load of resources, while deciding which strategy is the most promising to do filtering [13], has led to the successful elimination of these long lasting jobs. On the other hand, not each filtering job is the same, and taking also into account the properties of assigned jobs [15] has managed, in IR domain, to reduce the duration of filtering and to increase a user satisfaction.…”

“…Even though performed simulations [14] and real-time experiments [13] [15] prove that taking care of available resources and properties of jobs has brought significant improvements, a three year long authors' experience with many coordination schemes has also shown that the speed of learning has critical influence, as a filtering framework should improve itself during runtime. Unfortunately, upto-now invented coordination approaches are solely based on inductive learning, where training examples broaden knowledge about a given strategy.…”

“…Finally, it returns found results to U, and it hopes to get a relevance feedback that will be used as training data for further learning. While the remainder of this section is going to focus on the fitness initialisation, all other coordination steps are described in [13][14] [15].…”

“…Even though a particular strategy can be best suited for the assigned job, in a case where the needed system resource is highly loaded, that strategy will probably have weak chances to successfully produce the expected results. Self-improving coordination mechanism [13][14] avoids a selection of any strategy for which the unfavourable resource situation exists, and assigns a job to the strategy that will find results in a reasonable amount of time. This is simply achieved by estimating a current load of relevant system resources before making any decision about how every filtering strategy might be promising.…”