MRIB: Robust Task-Oriented Communication for Multi-Device Scenarios

In recent years, task-oriented communication has gained significant attention for its ability to extract and transmit necessary information for downstream tasks. However, multi-device collaboration scenarios pose challenges for task-oriented communication, such as encoding and fusing multimodal data and ensuring robustness in complex communication environ-ments. To address these challenges, we propose the Multi-device Robust Information Bottleneck (MRIB) scheme, which considers channel conditions in multi-device scenarios and introduces a new information bottleneck optimization objective. This objective is used to optimize the framework through end-to-end training and we derive a variational upper bound to facilitate the effective optimization of the framework. We validate the MRIB framework in both unimodal and multimodal scenarios, demonstrating its superiority over existing task-oriented communication methods, especially in environments with diverse and noisy data inputs. Experimental results show that MRIB offers better robustness and reasoning performance compared to baseline methods. Future work will focus on further optimizing the framework and exploring its applications in more complex communication environments.