Paul HONEINE

(Research Activities)

 

My Research World

My research journey is described here, through some milestones along the way, older works are at the end.


Feel free to have a look at some of my publications, here.


Have a nice journey

This work provides recent developments for testing stationarity of any signal relatively to an observation scale. The originality is to extract time-frequency features from a set of stationarized surrogate signals, and to use them for defining the null hypothesis of stationarity.
(more...)
PH-Stationarity_Testing.htmlPH-Stationarity_Testing.htmlshapeimage_5_link_0
This work extends the framework of the kernel learning machines to non-stationary signal analysis, by showing that some specific reproducing kernels allow these algorithms to operate in the time-frequency domain. This link offers new perspectives in the field of non-stationary signal analysis, which can benefit from the developments of pattern recognition and Statistical Learning Theory. (more...)
PH-TF_machines.htmlPH-TF_machines.htmlshapeimage_6_link_0
This work tackles the online identification problem for nonlinear and nonstationary systems using kernel methods. The order of the model is controlled by some sparsification criterion which leads to select the most relevant atoms to form a dictionary. We explore the dictionary adaptation using stochastic gradient descent methods along with an online kernel identification algorithm. The proposed method leads to a reduction of the instantaneous quadratic error and to a decrease in the model’s order.
(more...)
PH-Online_Adaptation.htmlPH-Online_Adaptation.htmlshapeimage_8_link_0
Kernel methods have been very useful to solve nonlinear identification problems. The main drawback of these methods resides in the fact that the number of elements of the kernel development, i.e., the size of the dictionary, increases with the number of input data, making the solution not suitable for online problems especially time series applications, adaptive filtering and online identification.
Our contributions in online learning with kernel machines are given here.
(more...)
PH-Online_Learning.htmlPH-Online_Learning.htmlshapeimage_9_link_0
Sensors are often randomly deployed, such as for monitoring inhospitable habitats and disaster areas, or equipping mobile entities, such as humans (e.g., Alzheimer patients in an adult daycare center) or animals in wilderness or even vehicles (in vehicular ad hoc networks (VANETs)).

Information captured by each sensor remains obsolete as long as it stays unaware of its location. Implementing a self-localization device, such as a global positioning system (GPS) receiver, at each sensor device may be too power-hungry for the desired application with battery-powered devices. Moreover, GPS or any global navigation satellite system (GNSS), are not functional in indoor environments.

The inference of the position of location-unaware sensors from inter-sensor ranging is a hard and ill-posed problem. We propose many contributions, including zoning, localization and tracking, with and without radio-location fingerprints. Investigated methods include machine learning, Bayesian methods, polar intervals, pre-image problem and matrix regression
(more...)
PH-Localization.htmlPH-Localization.htmlPH-Localization.htmlshapeimage_10_link_0
We propose machine learning techniques for collaborative learning in wireless sensor networks, taking into account the limited resources in computation, memory usage, communication as well as power. We are interested in wireless sensor networks deployed in an environment to sense for modeling and monitoring given physical phenomena, such as gas, pollution, biochemical, temperature, humidity, acoustic field, motion, etc. It is worth noting that collaborative learning greatly enhances the accuracy of the estimation, and the detection performance.
(more...)
PH-Wireless_Sensor_Networks.htmlPH-Wireless_Sensor_Networks.htmlshapeimage_11_link_0

Spectral Unmixing (Blind Source Separation) in Hyperspectral Image

This part of the website is under construction.

See references in the publications page, here.


Many nonlinear data processing, including kernel-based machine learning, operate a nonlinear map of the data to some feature space where linear processing techniques are operated. 
The pre-image problem considers the inference of the inverse map, namely the return from the feature space to the observation space. This is an ill-posed problem hard to solve. Our contributions include several techniques to solve the pre-image problem, and the extension of the scope to novel applications (beyond denoising). Our work opens the way to new classes of nonlinear methods in various research areas.
(more...)
PH-Pre-image.htmlPH-Pre-image.htmlshapeimage_12_link_0

Distributed Learning for Big Data

This part of the website is under construction.

See references in the publications page, here.


Online Detection with One-Class Classification

This part of the website is under construction.

See references in the publications page, here.