from October 10, 2026 to October 11, 2026
IA Cluster 3IA Côte d'Azur at the Fête de la Science 2026 in Antibes.
Discover the IA Cluster 3IA Côte d'Azur program for the Fête de la Science 2026 in Antibes!
- Date: October 10 & 11, 2026
- Times:
Saturday: 1:00 pm to 7:00 pm
Sunday: 10:00 am to 6:00 pm
- Place: Palais des Congrès d’Antibes (60 Chemin des Sables, Antibes)
SATURDAY, OCTOBER 10
1:00 pm to 6:00 pm
3D From Points to Objects: Teaching Machines to Understand 3D
Louis Hausseux (former IA Cluster 3IA Côte d’Azur PhD student)
How can we automatically distinguish a car, a tree, or a pedestrian in a point cloud?
Let’s discover how mathematics and artificial intelligence can teach machines to understand 3D environments.
A major prospect for robotics: combining reliable 3D perception with language models such as ChatGPT could pave the way for machines capable of understanding their environment, discussing it with us, and acting within it!
(Presentation inspired by my PhD research and the research perspectives of the Percolia project.)
3:00 pm to 3:45 pm
Artificial Intelligence and Applications in (Structural) Biology: What Are We Really Learning?
Conference by Frédéric Cazals (Researcher at Inria, Chairholder IA Cluster 3ia Côte d’Azur)
Proteins are essential building blocks of life, and understanding how they adopt their three-dimensional shape through a mechanism known as protein folding is crucial to understanding biological phenomena at the molecular level.
In this context, the 2024 Nobel Prize in Chemistry recognised AlphaFold, a program developed by DeepMind — a subsidiary of Google — which makes it possible to predict the three-dimensional structure of a protein from its amino acid sequence, leading to claims that the protein-folding problem had been solved.
This particular problem has brought considerable attention to the ability of artificial intelligence (AI) to make major contributions to science in general, and to biology in particular. But what do these generative AI models really teach us?
This talk will provide an opportunity to reflect on this question by exploring the connections between the concepts of a “solved problem”, knowledge, and know-how.
These are particularly important issues at a time when AI is raising complex questions about learning and the jobs of the future.
4:00 pm to 7:00 pm
Cardiac Digital Twin: Preventing Sudden Cardiac Death
Demo by Buntheng Ly (Postdoctoral Researcher at Inria, Epione team, alongside Chairholder Maxime Sermesant)
Sudden cardiac death is a major health concern in the developed world, reaching nearly 400,000 annual deaths in Europe. Ventricular arrhythmia is the phenomenon most frequently observed leading to sudden death, and thus robust and early detection of VA plays a major role in the management and prevention of SCD. In this presentation, we will detail the work carried out by our team (PhyBa, Inria) on cardiac digital twins for SCD management, covering the application of AI and cardiac modelling for VA prediction from non-invasive cardiac imaging.
SUNDAY, OCTOBER 11
10:00 am to 1:00 pm
Survival Stories: How to Teach Your Immune System to Fight Again
Giulia Foroni (IA Cluster 3IA Côte d’Azur PhD student, Olivier Humbert’s Chair)
Your immune system can already recognise and kill cancer cells, until a tumour learns to switch that recognition off. Survival Stories shows how doctors turn it back on, turning the real survival-analysis research behind that process into a hands-on science exhibit for all ages. Visitors can challenge their knowledge through a series of interactive games, from ranking the biological factors that matter most for a given treatment, to reading a survival curve, to stepping into a doctor's shoes to choose a treatment for a simulated patient with metastatic lung cancer.
1:00 pm to 4:00 pm
Cardiac Digital Twin: Preventing Sudden Cardiac Death
Demo by Buntheng Ly (Postdoctoral Researcher at Inria, Epione team, alongside Chairholder Maxime Sermesant)
Sudden cardiac death is a major health concern in the developed world, reaching nearly 400,000 annual deaths in Europe. Ventricular arrhythmia is the phenomenon most frequently observed leading to sudden death, and thus robust and early detection of VA plays a major role in the management and prevention of SCD. In this presentation, we will detail the work carried out by our team (PhyBa, Inria) on cardiac digital twins for SCD management, covering the application of AI and cardiac modelling for VA prediction from non-invasive cardiac imaging.
2:00 pm to 2:45 pm
Understanding Human Actions in Videos
Conference by François Bremond (Researcher at Inria, Chairholder IA Cluster 3ia Côte d’Azur)
Although generative AI has made tremendous progress in recent years, some areas remain to be explored. In this presentation, we will examine how human action recognition can be applied to the analysis of human behaviour captured through video streams and other sensors. Existing work has either focused on simple activities in real-life scenarios or on the recognition of more complex activities (in terms of visual variability) in manually selected video clips with clearly defined temporal boundaries. However, there is still a gap in methods capable of extracting events corresponding to complex human activities from continuous video streams in real-world scenes.
We will first present several techniques for detecting and tracking individuals in different environments. We will explore various modalities, such as skeleton recognition, optical flow, and emotion recognition, which can facilitate the activity recognition process. We will then review models from the literature for activity recognition and detection, including approaches based on self-attention, transformers, and different pre-training methods.
We will also discuss specific activity detection tasks, such as video anomaly detection using weakly supervised methods. We will then present new techniques for recognising Activities of Daily Living (ADLs) from videos. The proposed approaches to human activity analysis will be illustrated using datasets from “home care” applications, including Toyota SmartHome, NTU-RGB+D, Charades, and Northwestern UCLA.
This presentation will provide a better understanding of the current state of the art and future directions in action detection for human activity analysis across a variety of contexts.
4:00 pm to 6:00 pm
From 2D Images to 3D Variations: Editable Layer Decomposition and AI-Based Geometric Generation
Demo by Pierre Alliez (Researcher at Inria, Chairholder IA Cluster 3IA Côte d’Azur)
This demonstration presents an AI-assisted creation pipeline for transforming a 2D image into richly editable graphical and geometric content. It builds on two recent contributions: Illustrator’s Depth, which predicts a coherent layer order from a single image in order to decompose it into manipulable elements, and ShapeShifter, which interactively generates detailed 3D variations from a single reference model.
Starting from an illustration or photograph, the first step consists of estimating creative depth in the form of layer indices, providing an interpretable representation suitable for vectorization, local editing, and the generation of 3D reliefs. The second step uses these geometric representations to rapidly produce new 3D variations through a multi-scale diffusion model combining voxels, points, normals, and colors, while preserving the fine details of the original shape.
The demonstration highlights how these two approaches can be combined to move from 2D visual content to an interactive 3D design space, opening up new possibilities for graphic creation, generative design, digital fabrication, and next-generation content creation tools.
