
Dr. Patrice Wira
Patrice Wira is a Professor at IRIMAS (Institut de Recherche en Informatique, Mathématiques, Automatique et Signal) Laboratory, Université de Haute Alsace, Mulhouse, France. He received his M.S. degree and Ph.D. degree in electrical engineering from Université de Haute Alsace in 1997 and 2002, respectively. He has published several journal papers, book chapters and conference papers. He has also chaired and co-chaired numerous sessions and special sessions on intelligent control and machine learning techniques issues. His current research interests focus on artificial neural networks, adaptive control systems, neuro-control, and especially their applicability to robotics and to active power filtering. Pr. Dr. Wira was Deputy Director of IRIMAS from 2019 to 2020. Since 2020 he is Director of the University Institute of Technology of Mulhouse (IUT: Institut Universitaire de Technologie de Mulhouse). He is member of IEEE since 2004 and senior member of IEEE since 2013.
Artificial neural networks for exploring signals and data related to power and water consumption
Today, sensors can be installed anywhere to measure the environment and the uses of a building. The measured data can be centralized, processed and analyzed by algorithms in order to react quickly accordingly or to predict. The focus here is on energy, electricity and water consumption of one or more rooms or an entire building. The data are completed by external environmental data (temperature, pressure, luminosity, etc.) and internal data (door opening, human presence, etc.). These signals and consumptions directly reflect the activity of uses and are impacted by environmental data. These signals and data are very different in nature from each other and are difficult to be exploited together. Using traditional signal and data processing techniques alone is also not sufficient for a successful exploration. Although, the use of artificial neural networks machine learning (a set of methods/models capable of learning to solve a problem from the data) make it possible. This talk presents not only a comparison of different learning architectures but also provide insights about their improvements in analyzing and understanding these heterogeneous and big quantity of measures in closed environments occupied by users. Representations, primitives and features representative of the different groups of uses or users have been developed to recognize and identify them. Several supervised and unsupervised neural network approaches have been applied and individually tested for the classification of electrical devices according to the harmonic current characteristics they generate, for the prediction the next power and water consumption, for detecting abnormal consumption, etc. The proposed learning approaches have been validated on exploring real measured data and signals.

Dr. Messaoud HAMOUD
Messaoud HAMOUD is Professor at the University of Adrar . He received his diploma of Engineer and Magister in electrical engineering from The University of Sciences and Technology of Oran (USTO), Algeria. He obtained his Phd degree in electrotechnics in 2007 from The University of Sciences and Technology of Oran. Within the Unit of Renewable Energy Research in Saharan medium of Adrar, he has been managing director for seven years. . Currently, he is a Director of Laboratory of Sustainable Development and Informatics.The activities of this lab. focused on the new energy framework and its implications for the Distribution and Transmission Systems. Also,he is a member in many scientific and industrial commitees and director of several doctoral courses .Besides , plenty of conferences , meetings and study days ,both national and international , has been organized from his part. His research interests are optimization of energy flows in microgrids, the integration of renewable energies, HVDC systems, and the Electrical Discharges.
Energy transition and security in Algeria: What is it and why is it necessary?
This study deals with the issue of the transition towards renewable energies in Algeria, which is an urgent necessity embodied in strategies and programs for the gradual transition to the manufacture and generation of environmentally friendly renewable energies. The inevitable problem and the effectiveness of the transition towards renewable energies in Algeria includes several aspects, including the exploitation of renewable energies, which represents the real alternative to the aftermath of fossil fuels, especially with the exacerbation of its effects and biological risks and the serious and conscious thinking of the need for energy transition towards sustainable energies that do not exhaust its bidding and constitute a safe and vital source for sustainable development Without causing any harm to the ecosystem, It can also be traded across successive generations until the required continuity is achieved in sustainable development projects and justice in benefiting from intergenerational resources. It also includes strengthening Algeria's energy position in the international energy market. The importance of this topic lies in the fact that it deals with an important issue related to several aspects, whether political, economic, environmental, health, and security. In this topic, we will discuss Algeria's strategy in the energy transition towards renewable energies in the context of economic diversification, diversification of energy sources, and a sustainable development approach that takes into account the environment and future generations. Algeria has written an ambitious program in this regard, in which we address:
- Algeria's strategy in renewable energies
- Transition to a safer energy system and protection from industrial and nuclear risks.

Dr. İrem Ülkü
Remote sensing based semantic segmentation of trees
Increasing amounts of greenhouse gases, especially carbon dioxide (CO2), in the atmosphere has caused climate change, a fact that brings great difficulties of our times. In order to tackle the worst effects of global warming, forest management and monitoring have drawn significant attentions. Increasing trees is one of the best ways to slow down global heating by reducing the rate of CO2 in the atmosphere, for trees can effectively absorb carbon. Forest management is better achieved by means of tree detection, getting advantage of the remote sensing technology in the efficient collection of large data from remote and inaccessible landscapes. However, with the utilization of remote sensing technologies, there are still some challenges that cannot be accommodated even by high resolution satellites, such as color variations, occlusion, geometrical complexity and the difficulty of separating other types of green vegetation and small shrubs from trees. The main application in which the multiple spectral bands obtained from remote sensing prove to be extremely useful is the mathematical combination of them as vegetation indices. However using such hand-crafted vegetation indices alone is also not sufficient for a successful tree detection.
Although traditional machine learning techniques such as random forest (RF) and support vector machine (SVM) have been incorporated into the tree detection and segmentation by using remote imaging, the low-level extracted features cannot capture high-level features implicitly buried inside multispectral imagery. Deep learning is one of the most promising approaches for overcoming these challenges and semantic segmentation techniques developed within this rapidly evolving area are now state-of-the-art for many remote sensing based tree detection tasks. The potential to extract high-level, complex abstractions from multispectral data is a result of the deep learning model's ability to learn hierarchically. This talk will present not only a comparison of various different deep learning architectures such as UNet, SegNet, DLinkNet, and DeepLabv3+ but also provide insights about their performance improvements by fusing various vegetation indices or spectral bands for the aim of tree segmentation. This talk will also provide a better understanding of the several factors affecting the pixel-wise tree segmentation performances of deep learning models, such as spatial resolution, spectral resolution, input signal and decoder design.

Dr. Didem KETENOĞLU
Optimizing a Synchrotron Radiation Beamline by Genetic Algorithms: The GASOLINE Software
Since accelerator-based 3rd generation light sources (i.e. Synchrotron Radiation experimental beamlines) are unique tools in natural sciences research, optimization of miscellaneous components such as mirrors, lenses, monochromators as well as undulators, is a big concern for beamline scientists. In this respect, Genetic Algorithms provide realistic and reliable solutions when full beamline optimization is needed before commissioning. On this occasion, Genetic Algorithms enable a crucial double-check opportunity for the outcomes of dedicated optimization programs like xrtQook. Hence, the software “Genetic Algorithms based Synchrotron radiation Optimization for an X-ray beamLINE (GASOLINE)” has been developed to track and optimize radiation characteristics starting from the undulator until the experiment sample. In conclusion, the GASOLINE and xrtQook outcomes are comparatively discussed for a dedicated X-ray beamline.