In this to ExPEDite related paper, the authors introduce a reinforcement learning framework that uses partially defined rewards along with human feedback to generate energy-efficiency recommendations aimed at reducing energy costs. Their approach generalizes human preferences effectively, decreasing the reliance on detailed labelling from users while also enhancing the robustness of the learned reward model. The proposed method demonstrates improved performance over prior models by striking a balance between user input burden and recommendation quality in energy optimization tasks.

Read the full publication here: https://dl.acm.org/doi/full/10.1145/3679240.3734660

For a full overview of relevant publications visit this section on the ExPEDite website: https://expedite-project.eu/readings/interesting-readings-and-reports/