Projects

Decentralized high-dimensional statistical learning

Research in this context pursues the goal of understanding the overall communication, data, and computational trade-offs when solving decentralized (high-dimensional) statistical learning problems. Specific emphasis is put on the design of schemes that avoid complexities or communication costs scaling with the problem's ambient dimension.

Decentralized M-estimation: convex problems

Acceleration in Distributed Sparse Regression

M. Maros, G. Scutari, NeurIPS 2022

DGD2: A Linearly Convergent Distributed Algorithm for High-dimensional statistical recovery

M. Maros, G. Scutari, NeurIPS 2022

High-dimensional inference over networks: linear convergence and statistical guarantees

Y. Sun, M. Maros, G. Scutari, G. Cheng, submitted JMLR 2023.

Decentralized algorithms for sparse high-dimensional M-estimation

M. Maros, G. Scutari, submitted JMLR 2023.

A unified view of decentralized algorithms for sparse linear regression

M. Maros, G. Scutari, accepted to CAMSAP 2023.

Decentralized M-estimation: non-convex problems

Decentralized Matrix Sensing: Statistical Guarantees and Fast Convergence

M. Maros, G. Scutari, accepted to NeurIPS 2023.

Decentralized optimization and applications to Cyber-physical systems

Many applications naturally yield decentralized yet time-varying optimization problems, meaning some or all problem parameters become obsolete after a certain amount of time has passed. Research in this context pursues the goal of designing schemes that are robust to parameter changes, while providing theoretical guarantees that certify their performance.

Time-varying problems

Dynamic Power Allocation for Smart Grids via ADMM

M. Maros, J. Jaldén, IEEE SPAWC 2018.

ADMM for distributed dynamic beam-forming

M. Maros, J. Jaldén, IEEE Transactions of Information processing over Networks, 2018.

Time-varying networks

A dual linearly converging method for distributed optimization over time-varying undirected graphs

M. Maros, J. Jaldén, IEEE CDC, 2018.

ECO-PANDA: a computationally economic, geometrically converging optimization method on time-varying undirected graphs

M. Maros, J. Jaldén, IEEE ICASSP 2019.

A geometrically converging dual method for distributed optimization over time-varying graphs

M. Maros, J. Jaldén, IEEE Transactions on Automatic Control, 2020.