Hélène's Internship at ICHEC

Hélène de Foras, a recent graduate from Paris-Saclay University and ENSIIE (an Engineering School specialising in Computer Science), has spent the last six months at ICHEC completing her end-of-studies internship. Her work focused on optimising machine learning workflows, with a particular emphasis on semantic segmentation use cases. 

As part of an interdisciplinary team, Hélène addressed two major datasets: the HITL semantic segmentation dataset and FloodNet, which includes images of flooding from Hurricane Harvey. Utilising the powerful ICHEC supercomputer Kay and EuroHPC’s Meluxina, she executed her scripts and employed a range of profiling tools, such as cProfile, VizTracer, PyTorch Profiler, Scalene and Nsight Systems. Her efforts in profiling identified critical performance bottlenecks, leading to optimisations including multi-GPU support, enhanced data loading processes and efficient data distribution. 

To orchestrate the machine learning workflow, Hélène used TensorBoard and MLFlow, which facilitated effective job management and organisation. Her comprehensive documentation ensures that ICHEC collaborators can benefit from her findings and improvements. 

The internship also provided Hélène with the opportunity to attend three workshops on AI and supercomputing, further deepening her expertise. She also completed a Coursera certification in Deep Learning with PyTorch for Image Segmentation, complementing her practical experience. 

Expressing her gratitude, Hélène thanks the ICHEC team for their warm welcome and support, with special appreciation for her internship supervisor, Buket Benek Gursoy, for her invaluable guidance.