Thesis Defense - Aurora Rossi (Université Côte d'Azur, Centre Inria) - 25/09/2025
Keywords
Summary
148 words
Critical Evaluation
Value of the Information & Strength of the Argument
The presentation offers significant value by introducing novel computational methods for temporal network analysis, particularly the Random Temporal Hyperbolic null model and a machine learning framework with Shapley values for identifying task-relevant subnetworks. The argumentation is solid, supported by empirical results and comparisons with existing models. The speaker clearly explains the rationale behind each methodological choice and provides evidence for the effectiveness of the proposed approaches. The inclusion of open-source software and published datasets enhances the reproducibility and impact of the work.
Scientific Rigor, Source Quality, Title Accuracy
The scientific rigor is high, with a clear methodology, appropriate use of statistical tests, and validation through classification tasks. The sources cited include relevant literature and the speaker’s own publications, though specific references are not detailed in the talk. The title accurately reflects the content, and the presentation is well-structured. The talk includes a brief mention of collaborations and funding, but no explicit citations of external sources beyond the datasets and the GraphNeuralNetworks.jl package. The adequacy between title and content is strong.
179 words
Title / Content Match
The title accurately reflects the content: a thesis defense on computational methods for temporal networks with applications to neuroscience.
Quality & Reliability
8/10
The defense presents original research with clear methodology, published datasets, and a peer-reviewed paper. The speaker demonstrates deep knowledge and provides reproducible details. Minor limitations include lack of external validation and some presentation issues.
Key Moments
Markers derived by PSI from the transcript: the creator did not define chapters.
- Introduction to temporal networks and thesis overview
- Data pipeline: from raw fMRI to temporal networks
- Presentation of two published datasets
- Null models for significance testing
- Random Temporal Hyperbolic model and results
- Machine learning for narrative task classification
- Shapley values to identify important subnetworks
- Graph neural networks and GraphNeuralNetworks.jl
- Temporal graph convolution and applications
Cited Sources
- EUR DS4H funded theses — Funding information for the thesis
Concurring Sources
- GraphNeuralNetworks.jl — Open-source Julia package for graph neural networks, contributed to by the author.
Contribution & Novelties
The thesis contributes novel computational methods for temporal network analysis, including a new null model (Random Temporal Hyperbolic) and a machine learning framework with Shapley values for identifying task-relevant subnetworks. It also provides open-source datasets and software (GraphNeuralNetworks.jl) to facilitate research. The application to neuroscience offers new insights into brain dynamics during narrative comprehension.
Pour aller plus loin :
- Temporal networks — Overview of temporal networks.
- Graph neural networks — Introduction to GNNs.
- Shapley value — Explanation of Shapley values in game theory and XAI.
85 words
Radar Profile
The radar profile shows high scores in quantity and quality of information, with a strong technical level and good reliability. The presentation is dense and well-supported, indicating a solid scientific contribution.