We distinguish two complementary patterns, which we believe are representative of broad areas of data analytics:

1. Forwards ReComp. In this pattern, knowledge refresh decisions are triggered by changes that occur in the inputs to an analytics process, and are based on an assessment of the consequences of those changes on the current outcomes, in terms of expected value loss, or opportunities for value increase.
2. Backwards ReComp. Conversely, in this pattern the triggers are observations on the decay in the value of the outputs, and re-computation decisions are based on the expected value improvement following a refresh.

In both cases, when a limited re-computation budget is available, estimates of the cost of refresh are needed. Cost may be expressed, for instance, as time and/or cost of cloud resource allocation.

Next: The ReComp vision