Fair Learning for AI-RAN

AI-enabled Radio Access Networks (AI-RANs) promise to serve a wide range of users with time-varying learning needs, leveraging shared edge resources. A new study introduces an online-within-online fair multi-task learning (OWO-FMTL) framework to address the challenge of ensuring fair inference across different users.

OWO-FMTL: An Innovative Approach

OWO-FMTL uses two learning loops: an outer loop to update the shared model and an inner loop to rebalance user priorities. Equity is measured using a generalized alpha-fairness parameter, which allows balancing efficiency and fairness. The framework guarantees diminishing performance disparity over time and operates with low computational overhead, making it suitable for edge deployment.

Performance and Benefits

Experimental results, obtained on convex and deep learning tasks, demonstrate that OWO-FMTL outperforms existing multi-task learning baselines in dynamic scenarios. This approach offers an adaptive and lightweight mechanism to ensure fair resource allocation in AI-RAN environments.