Discover the 3IA Côte d'Azur program for the Fête de la Science 2024 in Antibes!
- Dates: October 12 and 13, 2024
- 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)
- All lectures, demos and presentations will be in French.
Saturday program
- 1:00 pm to 7:00 pm
-
- PEACE (Providing Explanations and Analysis for Combating Hate Expressions) by 3IA Techpool
This demonstrator describes a method of not only detecting explicit and implicit hate speech in text, but also explaining these predictions with natural language (using language models such as GPT).
- Learn3D by 3IA Techpool
- 2:00 pm to 3:00 pm
-
- Conference: "Recognizing human activity" by François Bremond, 3IA Chairholder
In this talk, we will discuss how Video Analytics can be applied to human monitoring using as input a video stream. Existing work has either focused on simple activities in real-life scenarios, or on the recognition of more complex (in terms of visual variabilities) activities in hand-clipped videos with well-defined temporal boundaries. We still lack methods that can retrieve multiple instances of complex human activity in a continuous video (untrimmed) flow of data in real-world settings.
Therefore, we will first review few existing activity recognition/detection algorithms. We will discuss various modalities, such as skeleton tracking, optical flow, gaze detection, and emotion recognition, that can aid in the activity recognition process. Then, we will present several novel techniques for the recognition of ADLs (Activities of Daily Living) from 2D video cameras. We will illustrate the proposed activity monitoring approaches through several home-care datasets: Toyota SmartHome, NTU-RGB+D, Charades and Northwestern UCLA. We will end the talk by presenting some results on medical (mental health), well-being and home-care applications. - 2:00 pm to 5:00 pm
-
- Can we trust AI in medical imaging by Vincenzo Marciano, 3IA PhD
The increasing reliance on AI in medical imaging has introduced significant advancements in diagnostic accuracy and treatment planning. However, once deployed, the effectiveness of these AI-driven models must be continuously validated to ensure their reliability in clinical settings. This raises the question: Can we trust AI in medical imaging? The need for rigorous quality control (QC) models is imperative to assess the performance of nowadays segmentation algorithms. These QC models must be capable of evaluating the accuracy and consistency of AI-generated segmentations, ensuring that they meet clinical standards.
- 2:00 pm to 6:00 pm
-
- A "green" robot for transfer operations: how to put digital and AI in their right places by Jean-Pierre Merlet, 3IA Chairholder, and Clara Thomas, 3IA PhD
Pick and place is the most common transfer operation performed by industrial robots. It is very simple, but currently requires a lot of energy due to the structure of the robotic arms and the computers used to control them.
Our aim is to create a new robot that consumes less energy to perform the Pick and Place operation. We want to change the structure of the robot (using cables) and control it electro-mechanically, without the need for a computer and reducing the use of electronic boards to a minimum. In this way, IT would be used upstream, during the design of mechanical parts, rather than during robot operation.
A first version of this "green" cabled robot, which will perform Pick and Place trajectories, will be presented.
Sunday program
- 10:00 am to 1:00 pm
-
- PEACE (Providing Explanations and Analysis for Combating Hate Expressions) by Serena Villata, 3IA Deputy Scientific Director and Chairholder
This demonstrator describes a method of not only detecting explicit and implicit hate speech in text, but also explaining these predictions with natural language (using language models such as GPT).
- 1:00 pm to 5:00 pm
-
- PEACE (Providing Explanations and Analysis for Combating Hate Expressions) by 3IA Techpool
This demonstrator describes a method of not only detecting explicit and implicit hate speech in text, but also explaining these predictions with natural language (using language models such as GPT).
- Learn3D by 3IA Techpool
This demo shows 3D shapes that have been reconstructed from point clouds using neural networks. This approach makes it possible to reconstitute particular shapes more effectively than is traditionally the case.
- 2:00 pm to 3:00 pm
-
- Conference: "Algorithms and learning for protein structure and function: biology and medicine at the molecular level" by Frédéric Cazals, 3IA Chairholder
The SARS-CoV-2 pandemic has highlighted the importance of understanding complex molecular mechanisms, in this case the infection of our cells by a virus, in order to combat it. The questions raised here concern the biophysical modeling of biomolecules, very large structures (several thousand or tens of thousands of atoms) in constant motion, and therefore very difficult to grasp.
This field, like many others, benefits from advances in Artificial Intelligence, with Google/Deepmind's AlphaFold2 software in particular, capable of predicting certain protein conformations in impressive fashion. AlphaFold2 has unequivocally launched a revolution in biological and medical research on a molecular scale.
This talk (for the general public) will take stock of these advances, evoking in turn the promise of these methods, but also their limitations, insofar as they still come up against a number of difficult mathematical and algorithmic problems in very high dimensions. - 2:00 pm to 5:00 pm
-
- Learnable representation of 3D shapes, or digital twins of construction sites by Pierre Alliez, 3IA Chairholder
Today's machine learning models are capable of creating results of many types using efficient numerical representations, such as pixels for images or words for text. However, despite a strong need for learning-based simulation, rendering or shape creation, there is currently no standard learnable representation for 3D shapes. Our aim is to bridge the gap between geometry processing and machine learning to make data-driven models compatible with such applications.