Thesis On Anomaly Detection in Quasi-periodic Time

Network monitoring using probes generates digital indicators at regular intervals that form time series characterized by a quasi-periodic behavior. It is impossible for operators to manually analyze the large number of recorded time series to detect possible anomalies, characterizing malfunctions. It is therefore necessary to facilitate their work by an automatic detection of these anomalies, which can appear simultaneously in distinct time series. It will be necessary to take into account this property by employing algorithms relative to multivariate time series.

One aspect to be taken into account is the difficulty of obtaining a sufficient base of examples for supervised learning, especially since the profile of the traffic curves is constantly evolving over time and they are also very different from one country to another.

Another important challenge for the telco is to detect the appearance of anomalies as quickly as possible in order to guide the diagnosis and undertake the appropriate corrective actions. Here again, it will be necessary to start by studying existing solutions [1].

The objective of this thesis is to propose unsupervised algorithms capable of real-time processing on high data rates, consisting of multivariate time series, which can automatically and continuously adapt to the evolution of the network. These algorithms should also be able to alert the operator in case the data leads them out of areas for which they are valid.

[1] B L Eckley I A Fisch, ATM. Real time anomaly detection and categorisation. Statistics and Computing, 32, 2022.

[2] F Liu et al. Anomaly detection in quasi-periodic time series based on automatic data segmentation and attentional lstm-cnn. IEEE Tr. on Knowledge and Data Eng., 2020.

[3] S Schmidl, et al. Anomaly detection in time series: a comprehensive evaluation. Proc. of the VLDB Endowment, 15 :1779-1797, 2022.

**about you**:

- Specialization in the field of statistics, machine learning, AI and/or signal processing
- Taste for applied mathematics and algorithmic development
- Good scientific programming skills: Python/Julia for prototyping and Rust for integration with Orange probe software;
- Fluency in English (written and oral). Knowledge of telecom networks would be a plus but is not essential.

**Desired experience (internships,...)**

Experience with a strong related component to the proposed problem will be an asset for this position (e.g. signal or time series processing, data analysis, unsupervised learning).

**additional information**:
For the scientific aspects, the combination of operational constraints (unsupervised learning, continuous automatic adaptation, real-time and on the fly detection) makes the anomaly detection problem a real scientific challenge. This context of anomaly detection combining unsupervised learning, real time and automatic adaptation is little addressed in the literature. The focus on quasi-periodic time series reinforces its original character.

For the applicative aspects, the work will be done on different profiles of real data, all from Orange network probes in operation. This will further anchor the problem on real uses and will also allow to orient the approaches to be explored.

The work will thus be systematically tested and valorized on the prototype probes developed by the team.

**department**:
Orange Innovation brings together the research and innovation activities and expertise of the Group's entities and countries. We work every day to ensure that Orange is recognized as an innovative operator by its customers and we create value for the Group and the Brand in each of our projects. With 720 researchers, thousands of marketers, developers, designers and data analysts, it is the expertise of our 6,000 employees that fuels this ambition every day.

Orange Innovation anticipates technological breakthroughs and supports the Group's countries and entities in making the best technological choices to meet the needs of our consumer and business customers.

Within the Innovation/Networks department, you will be part of a passionate team with a wide range of skills (IP network experts, developers, data scientists, real-time processing specialists); a team that specifies, develops and implements monitoring and diagnosis solutions in Orange networks that can be used directly by our operational teams to improve the Quality of Services and Networks, security and efficiency in the fight against fraud.

**contract**:
Thesis