Junior Telecommunications Engineer Specializing in AI and Signal Processing

Throughout my academic journey, I have developed a keen interest in artificial intelligence. This enthusiasm led me to undertake my second-year internship at IMS, focusing on the interpolation principle within the context of generative methods. This internship was part of societal advancements related to the development of new materials. Each material is characterized by microstructure attributes responsible for its mechanical properties. The objective of the internship was to generate, using a convolutional variational autoencoder, new interpolated images between the images in the database, representing novel mechanical properties. This would facilitate the design of new materials, referred to as virtual materials, as they are generated by algorithms rather than costly physical and chemical experimentation, which consumes time and human and financial resources. Being a cutting-edge topic, through a literature review primarily in English, this internship allowed me to develop a critical mindset and propose a solution to the problem posed within the three-month timeframe available to me.
 
 
Furthermore, the rise of machine learning has led to remarkable results, although often achieved at the expense of a considerable volume of data. Collaborating within a group of three students, I had the opportunity to work on a project focused on frugality in the field of supervised deep learning. To reduce the environmental impact of AI, models that are less resource-intensive have been highlighted. This can be beneficial in contexts where computing power, memory, or energy consumption are constraints, such as applications for smartwatches. The concept of frugality was explored in the project, focusing on three main axes: managing the database available to us, pruning unnecessary neurons, and implementing less costly calculations.
 

In addition to my engineering degree, I hold a research master's degree from the University of Bordeaux, focused on image restoration and high-resolution imaging. Building on this foundation, I completed a research-oriented final-year internship at the Institute for Research in Computer Science of Toulouse (IRIT). During this internship, I worked on ultrasound localization microscopy (ULM) using diffusion models, which are particularly well-suited for addressing inverse problems in an unsupervised manner. These approaches are especially relevant in the medical field, where ground truth data is often unavailable. The models were applied to super-resolution, enhancing image resolution, as well as to deconvolution, pushing beyond the physical limits of the ultrasound imaging system and compensating for the blurring introduced by it.
Beyond the specific case of ULM, which represents a particular application in medical imaging, the model was tested on a variety of ultrasound images. The promising results obtained have led to the drafting of a scientific article, which is currently being finalized and will be published soon. Additionally, a demonstration applied to photographic images is available on my GitHub repository: GitHub/anesgh58.