How Expectation propagation works part3(Machine Learning 2023) | by Monodeep Mukherjee | Nov, 2023

Photo by Leonsa on UnsplashDetection for Flat-Fading Channels primarily based on Expectation Propagation(arXiv)Author : Elisa Conti, Amina Piemontese, Giulio Colavolpe, Armando VannucciAbstract : This paper goals at tackling the issue of sign detection in flat-fading channels. In this context, receivers primarily based on the expectation propagation framework seem like very promising though presenting some vital points. We develop a brand new algorithm primarily based on this framework the place, not like earlier works, convergence is achieved after a single forward-backward move, with out extra inside detector iterations. The proposed message scheduling, along with novel changes of the approximating distributions’ parameters, permits to acquire important efficiency benefits with respect to the state-of-the-art resolution. Simulation outcomes present the applicability of this algorithm when sparser pilot configurations must be adopted and a substantial achieve in comparison with the present obtainable methods.2.Federated Learning as Variational Inference: A Scalable Expectation Propagation Approach (arXiv)Author : Han Guo, Philip Greengard, Hongyi Wang, Andrew Gelman, Yoon Kim, Eric P. XingAbstract : he canonical formulation of federated studying treats it as a distributed optimization drawback the place the mannequin parameters are optimized in opposition to a world loss operate that decomposes throughout shopper loss features. A current various formulation as a substitute treats federated studying as a distributed inference drawback, the place the objective is to deduce a world posterior from partitioned shopper knowledge (Al-Shedivat et al., 2021). This paper extends the inference view and describes a variational inference formulation of federated studying the place the objective is to discover a world variational posterior that well-approximates the true posterior. This naturally motivates an expectation propagation method to federated studying (FedEP), the place approximations to the worldwide posterior are iteratively refined by way of probabilistic message-passing between the central server and the purchasers. We conduct an intensive empirical examine throughout varied algorithmic issues and describe sensible methods for scaling up expectation propagation to the trendy federated setting. We apply FedEP on normal federated studying benchmarks and discover that it outperforms sturdy baselines when it comes to each convergence velocity and accuracy.

Recommended For You