☰

Loann GIOVANNANGELI

Associate Professor in Computer Science

LaBRI, University of Bordeaux, France

About me.

Since September 2024, I am Associate Professor in the Computer Science Department of the IUT (Technical School) in the University of Bordeaux, France. I am member of the BKB team of LaBRI institute where I conduct research on Information Visualization and Artificial Intelligence.

Since February 2024, I am a Postofoctoral fellow in the VisLab team (Pr. Huamin QU) of the Hong Kong University of Science and Technology (HKUST). My research focus is the AI4VIS field where I propose algorithms and methodologies dedicated to Information Visualization.

From 2020 to 2023, I conducted my Ph.D. at the University of Bordeaux (France) in the BKB team (``Bench to Knowledge and Beyond") of the LaBRI institute, under the supervision of Romain BOURQUI ; in close collaboration with Romain GIOT and David AUBER.

``Automatic Evaluation of Abstract Visualizations using Automated Learning Techniques" (in French)

From 1st October 2019 to 31st September 2020, I was Research Engineer in the BKB team of the LaBRI.

I obtained my Master degree in Computer Sciences (Software Engineering) in 2019 at the University of Bordeaux.

Research.

Keywords

Information Visualization, Deep Learning, Graph Visualisation, Automated Evaluation

Interests

My research interests are centered around Information Visualization (VIS) and Artificial Intelligence (AI) (Machine Learning [ML] and Deep Learning [DL]) and their mutual contributions: AI applications for VIS (AI4VIS), and VIS centered on AI applications (VIS4AI).

With the outbreak of IoT, the amount and complexity of collected data keeps increasing. This makes their efficient restitution even more important and many visualization techniques that yesterday enabled their exploration are now outdated. Hence, the VIS community keeps on defining new visualizations and guidelines that scale to these complex data.
In this context, the objective of my research is to propose new solutions, based on Deep Learning, to automatically generate and/or evaluate visualizations. For these two tasks, the heterogeneity of modern data makes it difficult to design efficient algorithms/protocols that can be generalized to any sample. On the other hand, DL techniques require large amounts of data to train and become efficient. Leveraging Deep Learning techniques in the context of VIS could enable to explore more exhaustively the parameters space of visualization techniques ; and define automated and reproducible criteria for VIS evaluations.

At the same time, Deep Learning models become increasingly complex. With this complexity arise several challenges such as understanding their behavior. Yet, understanding these model predictions is now necessary for several reasons. First, their democratization in many domains causes ethical issues. It becomes essential to understand how they make their predictions to increase transparency, trust and acceptability from the general public. The second reason is the explosion of the training cost (e.g., ecological, financial) of complex models. They often require to execute more operations for a sample, and they need to be trained on larger datasets. In addition, the number of hyper-parameters to tune is also increasing. Since model trainings are experimental explorations of a hyper-parameter space, it is mandatory to understand how the model works to efficiently drive this exploration. My interest consist in addressing this challenge by proposing explainable algorithms (XAI) and Visualization techniques to empower AI designers with efficient tools to design and understand their learning models.

Supervision

Publications.

Most PDF are accessible on https://hal.science/search/index/q/*/authIdHal_s/loanngio [loanngio]

C : Conference, score taken from ICORE conference portal
J : Journal, score taken from Scimago

J Q1 Overlap Removal by Stochastic Gradient Descent with(out) Shape Awareness Loann Giovannangeli, Frederic Lalanne, Romain Giot and Romain Bourqui IEEE Transactions on Visualization and Computer Graphics (TVCG) 2024 Extended version 10.1109/TVCG.2024.3351479
C A Guaranteed Visibility in Scatterplots with Tolerance Loann Giovannangeli, Frederic Lalanne, Romain Giot and Romain Bourqui IEEE Visualization (VIS 2023) Oct 2023 10.1109/TVCG.2023.3326596
J Q1 Toward Efficient Deep Learning for Graph Drawing (DL4GD) Loann Giovannangeli, Frederic Lalanne, David Auber, Romain Giot and Romain Bourqui IEEE Transactions on Visualization and Computer Graphics (TVCG) 2022 Extended version 10.1109/TVCG.2022.3222186
C A FORBID: Fast Overlap Removal By stochastic gradIent Descent for Graph Drawing Loann Giovannangeli, Frederic Lalanne, Romain Giot and Romain Bourqui International Symposium on Graph Drawing and Network Visualization (GD2022) 2022 10.1007/978-3-031-22203-0_6 Best Paper Award (Track 2)
J Q1 Color and Shape efficiency for outlier detection from automated to user evaluation Loann Giovannangeli, Romain Bourqui, Romain Giot and David Auber Visual Informatics 2022 10.1016/j.visinf.2022.03.001
C A Deep Neural Network for DrawiNg Networks, (DNN)² Loann Giovannangeli, Frederic Lalanne, D. Auber, Romain Giot and Romain Bourqui International Symposium on Graph Drawing and Network Visualization (GD2021) Sep, 2021 10.1007/978-3-030-92931-2_27
C B Analysis of Deep Neural Networks Correlations with Human Subjects on a Perception Task Loann Giovannangeli, Romain Giot, D. Auber, Jenny Benois-Pineau and Romain Bourqui International Conference Information Visualisation Jul, 2021 10.1109/IV53921.2021.00029
J Q2 Toward automatic comparison of visualization techniques: Application to graph visualization Loann Giovannangeli, Romain Bourqui, Romain Giot and David Auber Visual Informatics 2020 10.1016/j.visinf.2020.04.002

Teaching.

Master

IUT (Universitary Technical Institute)

Project supervision

Resources.

 [0000-0002-9395-6495]

 [loanngio]

 [Giovannangeli, L]