Over-the-Air Computation for Vertical Federated Learning

Abstract

Vertical federated learning (FL) is a critical tech-nology to support distributed artificial intelligence (AI) services in futuristic 6G systems, since it enables efficient and secure collaborative machine learning from a number of heteroge-neous devices in Internet of Things. In order to improve communication efficiency in vertical FL, we propose an over-the-air computation (AirComp) assisted vertical FL approach to achieve fast global aggregation. We theoretically establish the convergence analysis of the approach and thus propose to minimize the mean-squared error (MSE) of AirComp to reduce the optimality gap. So as to tackle the intractable non-convex problem, we propose an algorithm based on superiorization of bounded perturbation with convergence guarantee. Numerical experiments demonstrate that our proposed algorithm achieves low AirComp MSE in short running time, thereby improving the learning performance of vertical FL.