Cross-Evaluate Fairness and Energy Policies

Derive a cross-evaluation and scalarized fairness–energy Pareto analysis for PRISM policies trained separately to optimize the $\alpha$-fairness objective and the energy-efficiency objective.

Background

PRISM is evaluated in two separate settings: one policy is trained for the fairness-utilization objective and another for energy efficiency. The reported experiments do not test whether a policy optimized for one objective performs well on the other, nor do they characterize the trade-off between the two objectives through a joint scalarization or Pareto frontier. The authors explicitly leave both analyses unresolved.