Program > Program RFIAP

Monday

Invited Speaker 1 (Juliette Sénéchal) 9:15 - 10:15

AI Alignment Through EU AI Act

The launch of Chat GPT, a general-purpose generative AI system, on the global market in November 2022 has led to a proliferation of reflections on the new risks posed to humans by artificial intelligence systems. In particular, these reflections have highlighted the fact that generative AI, as a substitute for human language and human creativity, runs the risk of exposing humans to forms of cultural uniformisation and disinformation by deepfakes, undermining the ability of humans to compose a society and to adapt themselves to the constraints of their territory. In the context of these reflections, it is possible to observe a proliferation of soft and hard norms emanating from companies, governments on different continents, supranational organisations, researchers, etc., aimed at bringing AI systems into alignment with human values. This proliferation of norms raises three questions: what human values should be protected at a time when generative AI is taking a technological leap forward? What types of norms should be prioritised to implement this protection effectively: ethical norms, legal acts, technical standardisation? What territorial scope should these norms have? This contribution will study the normative approaches proposed by the future European regulation on artificial intelligence (the AI Act) in order to address these questions.

Session 1 (Computer vision) 10:45 - 12:00

Chair: Emilie Poisson Caillault

Invited Speaker 2 (Celestine Mendler-Dünner) 13:30 - 14:30

Performativity in Machine Learning

Algorithmic predictions create expectations, inform decisions and steer consumption. As such, predictions used in societal systems not only describe the population they aim to predict, but they have the power to change it; a prevalent phenomenon often neglected in theories and practices of machine learning. In this talk, I will start by introducing the calculus of performative prediction, that conceptualizes this phenomenon by allowing the predictive model to influence the distribution over future data. This dynamic perspective on prediction elucidates new solution concepts, optimization challenges, and brings forth interesting connections to concepts from causality and game theory that I will discuss. Moreover, it allows us to articulate two mechanisms fundamental to prediction, learning and steering, making explicit the important role of power in prediction, and giving us a means to measure it. I will end my talk on a discussion of collective action as a strategy to resist the power of digital platforms.

Session 2 (Reconstruction, meshes and 3D) 14:30 - 16:00

Chair: Guillaume Caron

Session 3 (Image modelisation) 16:30 - 17:45

Chair: Camille Kurtz

Session Poster 1 17:45 - 19:30

Tuesday

Prix de Thèse SSFAM/AFRIF (TBA) 9:00 - 10:15

Joint Session with CAp 1 10:45 - 12:00

Chair: Deise Santana Maia

Invited Speaker 3 (Matthieu Cord) 13:30 - 14:30

Vision and Language with transformers

Since the seminal publication on transformers in Natural Langage Processing, the Computer Vision community has rapidly adopted and adapted this model to a multitude of vision tasks. In this seminar, I will trace these major developments, starting with image classification. I will then explore various extensions, such as object detection, image segmentation and captioning. In particular, this will enable us to illustrate how transformers have revolutionized the modeling of Vision and Language relationships.

Session 4 (Biomedical imaging) 14:30 - 16:00

Chair: François Rousseau

Session 5 (Image modeling, processing and analysis) 16:30 - 17:45

Chair: Nicolas Passat

Assemblée Générale 17:45 - 18:30

Session Poster 2 18:30 - 19:30

Wednesday

Invited Speaker 4 (Xavier Alameda-Pineda) 9:00 - 10:00

Learning for Companion Robots: Preparation and Adaptation

Companion and social robots are a chimeric mirage. On the one side, they are part of our collective representation of ready-to-deploy technological developments. On the other side, there are a series of scientific and technological barriers to such deployment. One of the scientific challenges is the difficulty of such robots to successfully perform in the wide variety of environments and social situations they need to face. In this talk, a series of works developed at the RobotLearn team at Inria at Univ. Grenoble Alpes will be presented, all aligned with the philosophy of developing learning strategies that would allow a social/companion robotic platform to quickly adapt its perception and action models to unseen situations. Applications such as multi-person tracking, audio-visual speech modeling, and navigation policies will be discussed.

Joint Session with CAp 2 10:30 - 12:00

RFIAP HAL collection