Website Building Software
We are continuously recruiting Postdoctoral researchers, PhD candidates, and MSc students in Machine Learning, Computer Science, Electrical Engineering, and related fields.
Interested candidates are welcome to contact me at any time throughout the year.
The list of announcements below may be incomplete.
Research topic
Change-point detection (CPD) is a fundamental problem in statistics and machine learning, focused on identifying abrupt shifts in the properties of data over time. These shifts, known as change-points, indicate transitions in the underlying distribution or dynamics of a system, which may result from external events or internal structural changes. The objective of CPD is to pinpoint when these changes occur and, in some cases, to understand the nature of the shifts. CPD has a wide range of applications across domains that require online insights and adaptive decision-making, such as medical monitoring, real-time trading, and network security. A growing number of these applications are generating structured, high-dimensional data with non-trivial and intricate geometric properties. These data often display complex relationships and dependencies that go beyond Euclidean spaces, necessitating sophisticated techniques for analysis and interpretation. Prominent examples include time sequences on groups and manifolds, time sequences of graphs, graph signals, etc.
A major challenge with dynamic structured data is finding representations that can effectively handle their underlying geometry, which is often defined by application-specific pseudo-distances. A common approach is to embed such data into conventional geometric spaces, like Euclidean spaces, even when the data may be more naturally represented in non Euclidean domains. This mismatch in representation complicates both learning and inference processes. Another challenge is that dynamic structured data are generated by a variety of sensors and infrastructures that continuously produce, disseminate, and store information. However, this data deluge already surpasses our capacity for analysis and decision making, necessitating online actions rather than offline processing due to its time-sensitive nature. The monitoring of telecommunications and energy production and distribution networks are characteristic examples of such time-critical applications.
The project aims to propose unsupervised online CPD algorithms for dynamic structured data, with a particular focus on time sequences on groups and manifolds. Non-parametric frameworks will be specifically considered, as they make fewer, if any, assumptions about the data’s underlying distribution and are better suited for detecting a wider variety of changes. The CPD algorithms will be designed to be computationally efficient to ensure scalability. Finally, an open-world recognition setting will be developed to classify changes on-the-fly into either previously seen classes or unknown classes. Applications to smart cities monitoring are considered.
Required Qualifications and Skills
A master’s degree in a field related to machine learning, such as computer science, electrical engineering, mathematics, statistics, or physical science.
Strong background in applied mathematics and statistics. Additional qualifications in optimization and numerical methods would be appreciated.
Strong programming skills in languages such as Python, C++ or MATLAB
To prepare your application, please send me an email
The next application step will include:
• Curriculum vitæ including the list of the scientific publications
• Motivation letter
• Letter of recommendation of the Master thesis supervisor
Localisation
Campus Valrose, Université Côte d'Azur, Nice (downtown), France
Contact
Cédric Richard (cedric.richard@univ-cotedazur.fr)
Research topic
Optical fiber, in addition to being a means of transmitting information, is also a material that is very sensitive to environmental variations. When a laser light pulse travels through an optical fiber, it interacts with tiny impurities in the material, causing optical backscattering. Processing this response provides estimates of the local variations in acoustic pressure along the fiber, over distances ranging from 40 km to 140 km with some systems. This technique, called Distributed Acoustic Sensing (DAS), is currently experiencing growing interest in an increasing number of applications, such as traffic monitoring, structural health monitoring, and natural hazard detection, to name a few. As telecom fibers are ubiquitous in urban environments, DAS represents a breakthrough concept to upgrade existing fiber optic networks to acoustic sensor arrays, becoming a key component for managing smart cities.
Except for a few applications, DAS data are typically processed on a trace-by-trace basis, essentially treating the instrument as a collection of conventional seismometers. Since these processing routines do not leverage the spatio-temporal coherence of DAS data, they do not reach their full potential. Recently, through the use of Riemannian geometry, we proposed a nonparametric framework for change-point detection in manifold-valued data streams. We applied these methods to detect micro-seismic events in borehole and underwater DAS data based on their long-range coherence rather than their energy. Although they achieved superior performance and are built on stochastic Riemannian optimization algorithms, these methods still suffer from limitations in computational complexity.
The post-doctoral fellow will build upon this preliminary work to investigate the development of more efficient online learning algorithms for manifold-valued data streams, with an initial focus on change-point detection, opening the door to new unsupervised data exploration methods. Next, the post-doctoral fellow will consider designing distributed learning algorithms for streaming manifold-valued data. Experiments will be carried out on urban, coastal, and underwater DAS data. The novelty of the application and the relative lack of a framework for processing DAS data should ensure the fast dissemination of this work.
Required Qualifications and Skills
A PhD degree in a field related to machine learning, such as computer science, electrical engineering, mathematics, statistics, or physical science.
Strong background in applied mathematics and statistics. Additional qualifications in optimization and numerical methods would be appreciated.
Strong programming skills in languages such as Python, C++ or MATLAB
To prepare your application, please send me an email
The next application step will include:
• Curriculum vitæ including the list of the scientific publications
• Motivation letter
• Letter of recommendation of the PhD thesis supervisor
Localisation
Campus Valrose, Université Côte d'Azur, Nice (downtown), France
Contact
Cédric Richard (cedric.richard@univ-cotedazur.fr)
Website Building Software