Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Filters








2 Hits in 4.6 sec

GLASU: A Communication-Efficient Algorithm for Federated Learning with Vertically Distributed Graph Data [article]

Xinwei Zhang, Mingyi Hong, Jie Chen
2023 arXiv   pre-print
For graph-structured data, graph neural networks (GNNs) are competitive machine learning models, but a naive implementation in the VFL setting causes a significant communication overhead.  ...  Vertical federated learning (VFL) is a distributed learning paradigm, where computing clients collectively train a model based on the partial features of the same set of samples they possess.  ...  Conclusion We have presented a flexible model splitting approach for VFL with vertically distributed graph data and proposed a communication-efficient algorithm, GLASU, to train the resulting GNN.  ... 
arXiv:2303.09531v1 fatcat:4z2quugatzdxlfs6hbwmlkbhf4

Interconnection Network Analysis and Design An Adaptable Parallel Search of Knovvledge Bases with Beam Search Survey of the MAP Project Realibility Prediction of Parsys Hypercube Architecture Komunikacija človek-računalnik v regulacijski tehniki Nevronske mreže Rekurzivni postopek testiranja večnivoiskesa komunikacijskega

Špigel, A Zeleznikar, A Žagar, P Brajak, S Prešem, P Brajak, L Vogel, A Zeleznikar, M Colnarič, I Rozman, R Piskar, M Debevc (+5 others)
unpublished
Acknowledgments I vvish to thank ali my coUeagues at the Department,for the Study of Intelligent Systems at OZIR for many hours of inspiring discussion and a fevv years of Al research, especially Ivan  ...  I vvould certainly like to see stronger biological support for this vievv, vvhich I nevertheless consider intuitive enough to be convincing.  ...  The mapping is reguired when the communication graph of a parallel algorithm differs from the interconnection architecture of the physical parallel machine.  ... 
fatcat:zp63d6tt4nehrm7jnafyshuynq