A quick and relatively simple way to accomplish this is to use a linear feature selection process [17]. The designer picks a set of N candidate attributes based on expert intuition. The first step involves simulating N schedulers, each of which uses only one of the N candidate attributes to determine the state of the memory system. Among these N attributes, the designer picks the attribute t1 that optimizes an objective function (e.g., performance). Then, the designer repeats the selection process with N - 1 schedulers, each one considering t1 and one of the remaining N - 1 attributes. After I ≪ N iterations, the process concludes, and the i attributes picked determine the state representation. set of state attributes.