This paper introduces a novel optimization framework for learning time-varying network topologies and imputing missing data from incomplete graph signals. The proposed method integrates graph and signal recovery, enhancing robustness, especially when data is sparse. It utilizes a fused-lasso regularizer to promote temporal smoothness in network dynamics and employs an efficient Proximal Alternating Direction Method of Multipliers (PADMM) algorithm for scalability. The authors provide theoretical convergence guarantees and non-asymptotic statistical error bounds, demonstrating superior performance over existing methods in numerical experiments. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Presents a new method for analyzing dynamic network data, potentially improving applications in fields reliant on time-series graph analysis.
RANK_REASON This is a research paper published on arXiv detailing a new methodology for graph learning and signal imputation. [lever_c_demoted from research: ic=1 ai=1.0]