Alexis Alzuria

Participation in the ANR DeepECG4U project, aiming to apply deep learning to the screening of cardiovascular pathologies from ECG signals.

Development and evaluation of interpretability methods (XAI) to improve the transparency and reliability of predictive models in the medical field.

Design of constrained approaches integrating medical concepts (heart rate, leads, specific anomalies) to enhance the clinical consistency of predictions.

Design of Python tools for the analysis of multispectral images of Porphyra, enabling automatic identification of areas with high growth potential. Use of GIS (Geographic Information Systems) for spatial processing and mapping of optimal areas. Contribution to field campaigns for the collection and validation of multispectral data. Development of a PyQt5 graphical interface as a plugin within GIS software to automate processing and facilitate analysis by researchers unfamiliar with programming languages. Results: optimal detection, accelerating crop planning.

In-depth analysis of the limitations of the Ward's method implementation in scipy.cluster.hierarchy. Design and development of an alternative linkage function minimizing the increase in intra-group inertia for better cluster coherence.

Implementation of alternative Python modules to those in scikit-learn, validated theoretically and experimentally. Production of a reusable Python library for future laboratory students, facilitating research reproducibility and pedagogical experimentation. Results: slight improvement in grouping stability and operational transfer of the solution for use in teaching and research.