Paul Irofti

About me: 
 Resume (RO)
 Security Seminar


 Sisteme de Operare
 Utilizarea SO
 OS Security
 Vedere Artificială
 Static Analysis
 Prelucrarea Semnalelor
 Calcul Numeric

 [E-mail address]

DDNET – Data Driven Fault Accommodation for Distribution Networks






Paul Irofti -- Principal Investigator
Florin Stoican -- Mentor


[1] P. Irofti, F. Stoican, and V. Puig, “Fault Handling in Large Water Networks with Online Dictionary Learning,” Journal of Process Control, vol. 94, pp. 46--57, 2020. [ bib | DOI | http ]
[2] A. Pătrașcu and P. Irofti, “Stochastic proximal splitting algorithm for composite minimization,” Optimization Letters, pp. 1--19, 2021. [ bib | DOI | .pdf ]
[3] C. Rusu and P. Irofti, “Efficient and Parallel Separable Dictionary Learning,” in Proceedings of the IEEE 2021 27th International Conference on Parallel and Distributed Systems (ICPADS). 2021, pp. 1--6, IEEE Computer Society. [ bib | http ]
[4] A. Pătrașcu and P. Irofti, “Computational complexity of Inexact Proximal Point Algorithm for Convex Optimization under Holderian Growth,” pp. 1--42, 2021. [ bib | arXiv ]
[5] P. Irofti, L. Romero-Ben, F. Stoican, and V. Puig, “Data-driven Leak Localization in Water Distribution Networks via Dictionary Learning and Graph-based Interpolation,” 2021, pp. 1--6. [ bib | arXiv ]
[6] P. Irofti, C. Rusu, and A. Pătrașcu, “Dictionary Learning with Uniform Sparse Representations for Anomaly Detection,” 2021, pp. 1--6. [ bib | arXiv ]
[7] P. Irofti, A. Pătrașcu, and A.I. Hîji, “Unsupervised Abnormal Traffic Detection through Topological Flow Analysis,” in 2022 14th International Conference on Communications (COMM). 2022, pp. 1--6, IEEE. [ bib | DOI | http ]
[8] A. Pătrașcu and P. Irofti, “On finite termination of an inexact Proximal Point algorithm,” Applied Mathematics Letters, vol. 134, pp. 108348, 2022. [ bib | DOI | http ]



The main goal of this project, called DDNET, is to adapt and propose new dictionary learning methods for solving untractable fault detection and isolation problems found in distribution networks. Given a large dataset of sensor measurements from the distribution network, the dictionary learning algorithms should be able to produce the subset of network nodes where faults exist.

Since a model-based approach is impractical, we consider here the data-driven alternative where the data is provided by the network sensors and processed as signals laying on a graph. Graph signal processing is a new and very active field where data-driven methods such as sparse representations with dictionary learning have shown promising results.

Sparse representations are linear combinations of a few vectors (named atoms) from an overcomplete basis (called dictionary). Formulating sensor placement as a graph sparse representation problem and modeling large-scale utility networks via a dictionary that is trained from sensor data, was attempted only very recently and as far as we are aware only by our team members.

The project objectives are to provide sparse modeling for sensor placement and to perform and improve data-driven fault detection and isolation with dictionary learning through multi-parametric reformulations of the standard algorithms while exploiting the underlying geometrical and topological structure of the data.

The research team covers the positions of principal investigator Paul Irofti with extensive expertise in dictionary learning and his mentor Florin Stoican, fault tolerant control expert. The team had a fruitful collaboration in the past during two national research projects.