An arXiv preprint proposes FLICKER, a federated spectral estimation algorithm for settings where a large matrix is horizontally partitioned across servers that cannot share raw data. The method uses a target-specific, gradient-based power iteration to recover a singular component from the distributed matrix.

FLICKER is presented in the context of community detection in bipartite networks. By keeping raw data in place, the algorithm aims to support privacy-preserving analysis while still allowing servers to collaborate on a common spectral estimate.

The abstract describes the algorithmic design but does not report experimental results or comparisons with other federated approaches. The contribution is a way to steer power iteration toward a specific target in a federated setting, rather than requiring centralized data or generic iterative methods.