In this paper we propose to follow an automated approach to solve this problem. Specifically, we devise a genetic algorithm (GA) [23] to explore the search space of possible reward functions for a given objective function. 4 Genetic algorithms (GAs) [23] are a heuristic search technique that is based on evolutionary processes. GA starts by randomly generating a population space of individuals,where each individual is a candidate solution for the problem being solved. Typically individuals in the population set are represented in binary as strings of 0’s and 1’s, but other encodings are also possible. Evolution is performed in generations and it starts by evaluating the fitness of the initial population according to an objective function. (Evaluation means simulations of candidate DRAM schedulers in our case.) Based on the fitness of individuals in the population set, the next generation of individuals are determined stochastically using some form of fitness based selection technique. These new individuals are then further evolved using operations like crossover and mutation, which leaves us with a population for the next generation. This is done iteratively until a certain number of generations has been evolved, or when a certain fitness level has been reached, after which the search is terminated. While many of the individuals in the initial population might not do anything useful, the evolutionary nature of GAs allow some of them to evolve into meaningful, high-performing solutions, and shed the rest in the process.