Development of federated learning neural networks with combined horizontal and vertical data partitioning
Year: 2026
Type: article
Abstract: Federated learning trains a global machine learning model without moving the distributed data, preserving their privacy. Most of the existing works are on either horizontal- or vertical-federated learning. This paper proposes two new federated learning frameworks for different combined horizontal and vertical data-partitioned scenarios based on neural networks. The data is horizontally partitioned for a subset of clients, labelled as horizontal clients, it is further vertically partitioned with another client, labelled as vertical client. In the first proposed framework, referred to as ‘Horizontal-OutputFed’, the output feature is associated with horizontal clients, while in the second proposed framework, referred to as ‘Vertical-OutputFed’, the output feature is associated with vertical client. The accuracy, loss and convergence rate of the Horizontal-OutFed are the same as achieved with a horizontal partitioned federated learning framework. Further, the impact of non-identically and independently distributed data is the same in both the Horizontal-OutFed and the horizontal frameworks for combined features. Compared to a centrally trained model, the globally trained model in the VerticalOutFed has the same performance metrics such as accuracy and convergence rate. Moreover, there is no effect of imbalanced data on the performance of the Vertical-OutFed, contrary to some of the previous work on individual horizontal- or vertical-federated learning. Privacy attack analysis for active, passive, white-box, and black-box attacks demonstrated that there was no risk of the leakage of the input and output features, nor could any client infer the data from resulting models or while sharing hidden layers’ outputs.
Source: Applied Soft Computing
Authors: Amir Anees, Matthew Field, Lois Holloway
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,730
Funder: Australian Research Data Commons
OpenAlex ID: W7127161925