Generating Schedules to Maximize Quality

In knowledge generating production processes such as intelligence gathering and news reporting, the quality of the result produced by a given activity depends on its duration, and, due to resource limitations and process deadlines, tradeoffs must invariably be made regarding how much time to devote to various activities to achieve maximum overall effect. In essense, some activities must be executed in faster time cycles than would be desirable under non-constraining circumstances, with consequent degradation of process quality. In this paper, we consider this type of scheduling problem, which we refer to generally as quality maximization. Starting with normal resource-constrained project scheduling problem (RCPSP) assumptions, we define a new type of scheduling problem by additionally associating a quality profile with each activity in the project. Quality profiles have an “anytime” property, implying that activities can be terminated at any point, with the quality of the output being proportional to duration. Instead of finishing all activities as fast as possible, the goal of scheduling in this context is to maximize the overall quality given a hard project due date. We formulate this quality maximization problem as a constraint-based optimization problem, and present a new constraint posting algorithm for solving this problem that incorporates a linear optimization program. Different constraint posting heuristics are defined and evaluated on a set of quality maximization RCPSP problems constructed from standard reference RCPSP problems. The experimental results show the overall effectiveness of our approach for generating schedules to maximize quality. And ratio-based heuristics provide a promising starting point for stochastic sampling or other schedule refining techniques by solving 100% problems without backtracking.