New paper on carbon-aware distributed inference

Our NSF-sponsored research, CADI, was presented at IEEE COINS in Madison, Wisconsin. This work explores distributing inference across a source and a partner edge device connected to different power grids, aiming to minimize the overall carbon footprint. It introduces a carbon footprint model for distributed inference that integrates computation, communication, and the hold time of partnering edge devices. An optimization formulation is proposed to determine how inference tasks should be distributed across these devices based on the carbon footprint model. Congratulations to the first author Robert Viramontes.