INRIA Grenoble PhD Seminars

November 2026

Speakers :

• Razan Mhanna, PhD student at STATIFY team (Inria) : Brain Graphs or Raw Signals? Rethinking Machine Learning for rs-fMRI
  Website LinkedIn Google Scholar
   Resting-state functional magnetic resonance imaging (rs-fMRI) offers a non-invasive tool for studying spontaneous brain activity in healthy individuals and in people with conditions such as autism, attention-deficit/hyperactivity disorder, and disorders of consciousness. Regional fMRI signals are commonly summarized as functional connectivity graphs, in which nodes represent brain regions and edge weights quantify pairwise statistical correlation between their time series. Although these representations facilitate learning from brain networks, they may discard signal properties relevant to prediction.
   In this talk, I will present two complementary studies examining how data representation and modeling choices affect rs-fMRI prediction. First, I will introduce a benchmark comparing three families of learning methods—classical machine learning, graph kernels, and graph neural networks—across four rs-fMRI datasets, evaluating both predictive performance and computational cost. Second, I will explore learning directly from regional time series, using controlled experiments with temporal shuffling and Fourier phase randomization to assess how preserving or disrupting specific signal properties affects predictive performance.
   Together, these studies offer complementary perspectives on connectivity-based representations and direct time-series learning, highlighting the importance of evaluating data representations, learning algorithms, and experimental protocols jointly.
1. Mhanna, R., Achard, S., Petersen, A., & Richiardi, J. (2026). A comprehensive benchmark of graph neural networks, graph kernels, and classical machine learning approaches on rs-fMRI brain graphs. In Medical Imaging with Deep Learning-Validation Papers.
2. Mhanna, R., Aykut, Y., Petersen, A., Richiardi, J., & Achard, S. (2026). Beyond functional connectivity estimation: Evaluating direct learning from rs-fMRI time series. hal-05851784


• Suraj Gupta, PhD student at ARMADHA team (Inria)
   TBA