Grants
StOpAnomaly - Algoritmi stohastici de optimizare pentru detectie de anomalii (Stochastic Optimization Algorithms for Anomaly Detection)
A grant of the Romanian Ministry of Education and Research, CNCS - UEFISCDI
Contract number: PD 143 din 20/08/2020
Code: PN-III-P1-1.1-PD-2019-1123 within PNCDI III
Team: Andrei Pătrașcu, Bogdan Dumitrescu
With the ever increasing amounts of data around us, many modern technologies designed for anomaly detection, computer vision or pattern recognition are now routinely using machine learning and data mining techniques. Anomaly detection is the problem of distinguishing between normal data samples with well designed patterns or signatures and those that do not conform to the expected profiles. Unsupervised classification and decomposition models shows more and more success on fraud or intrusion detection applications as it is emphasized by valuable literature sectors. Promising research directions in this branch move towards eliminating the influence of outliers from predictor performance, or towards inclusion of additional descriptions of data or relational information about data that could boost the model performance. However, since the high quality models based on these new features would require complicated stochastic optimization formulations, in StOpAnomaly project we aim to create, analyze and implement stochastic splitting algorithms for composite optimization problems with focus on robust anomaly detection models based on decomposition and one-class classification, empowered with background information. On short, the outlier influence will be minimized thorugh finding specific robustness-promoting loss functions and the background information synthesis will be approached by Privileged Information or Graph Embedding.
Objectives:
1. Development of stochastic first-order methods for (non)convex minimization
2. Acceleration techniques for large-scale problems
3. Application to one-class classification models
4. Application to dictionary learning models
5. Implementation and evaluation of the algorithms
Technical reports:
Papers:
A. Patrascu, P. Irofti, Stochastic proximal splitting algorithm for composite minimization, Optimization Letters, 15, 2255–2273, 2021. https://doi.org/10.1007/ s11590-021-01702-7.
A. Patrascu and P. Irofti, On finite termination of an inexact Proximal Point algorithm, Applied Mathematics Letters, 2022, https://doi.org/10.1016/j.aml.2022.108348.
P. Irofti, C. Rusu, A. Patrascu, Dictionary Learning with Uniform Sparse Representations for Anomaly Detection, ICASSP, 2022. (pdf)
P. Irofti, A. Patrascu, A. I. Hiji, Unsupervised Abnormal Traffic Detection through Topological Flow Analysis, 14th International Conference on Communications (COMM), 2022, 10.1109/COMM54429.2022.9817285
A. Patrascu, Finite convergence of the inexact proximal gradient method to sharp minima, submitted, 2021. (pdf)
A. Patrascu, P. Irofti, Complexity of Inexact Proximal Point Algorithm for minimizing convex functions with Holderian Growth, submitted, 2021 (pdf).
Preprints:
A. Patrascu, Block-Alternating Minimization Algorithms for Trimmed Optimization Problems, preprint, 2022.
A. Patrascu, C. Paduraru, Stochastic proximal algorithms for sharp convex functions: application to anomaly detection in graph signals, preprint, 2022.
A. Patrascu, P. Irofti, Truncated convex models for robust learning, preprint, 2022.
Software:
Splitting Proximal Algorithms - Python package that implements stochastic and deterministic Proximal (Sub)Gradient Methods for convex composite optimization.