Adrien Bardes — AMI Labs

Biography

Adrien Bardes is a Member of Technical Staff at AMI Labs, specializing in artificial intelligence, machine learning and computer vision. His research explores how AI systems can learn useful representations directly from images and video, reflecting his belief that visual understanding will be central to machines that can reason about the world.

Before joining AMI Labs, Bardes was a research scientist at Meta and conducted his doctoral research jointly with Meta AI and Inria’s WILLOW team under Jean Ponce and Yann LeCun. He is best known for co-creating VICReg, a widely influential method for preventing representation collapse in self-supervised learning, and for subsequent work including VICRegL, which learns local and global visual features simultaneously, and MC-JEPA, which combines the learning of visual content and motion within a joint-embedding predictive architecture.

Bardes earned a PhD in mathematics and computer science from École Normale Supérieure and completed the Mathematics, Vision and Learning master’s program at ENS Paris-Saclay. His earlier research experience included work at Inria and Carnegie Mellon University’s Robotics Institute, where he studied few-shot learning in computer vision.

Career History

2024-2026

Meta

Research Scientist

2020-2024

Facebook AI

Research Assistant, PhD

Key Papers

MC-JEPA: A Joint-Embedding Predictive Architecture for Self-Supervised Learning of Motion and Content Features

Self-supervised learning of visual representations has been focusing on learning content features, which do not capture object motion or location, and focus on identifying and differentiating objects in images and videos. On the other hand, optical flow estimation is a task that does not involve understanding the content of the images on which it is estimated. We unify the two approaches and introduce MC-JEPA, a joint-embedding predictive architecture and self-supervised learning approach to jointly learn optical flow and content features within a shared encoder, demonstrating that the two associated objectives; the optical flow estimation objective and the self-supervised learning objective; benefit from each other and thus learn content features that incorporate motion information. The proposed approach achieves performance on-par with existing unsupervised optical flow benchmarks, as well as with common self-supervised learning approaches on downstream tasks such as semantic segmentation of images and videos.

2023 · arXivLabs

64 citations

Vicreg: Variance-invariance-covariance regularization for self-supervised learning

Self-supervised learning of visual representations has been focusing on learning content features, which do not capture object motion or location, and focus on identifying and differentiating objects in images and videos. On the other hand, optical flow estimation is a task that does not involve understanding the content of the images on which it is estimated. We unify the two approaches and introduce MC-JEPA, a joint-embedding predictive architecture and self-supervised learning approach to jointly learn optical flow and content features within a shared encoder, demonstrating that the two associated objectives; the optical flow estimation objective and the self-supervised learning objective; benefit from each other and thus learn content features that incorporate motion information. The proposed approach achieves performance on-par with existing unsupervised optical flow benchmarks, as well as with common self-supervised learning approaches on downstream tasks such as semantic segmentation of images and videos.

2021 · arXivLabs

2,129 citations

More Publications (top 5 by citations)

V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning

2026 · arXiv.org

47 citations

Hierarchical Planning with Latent World Models

2026 · arXiv.org

12 citations

MJEPA: A Simple and Scalable Joint-Embedding Predictive Architecture for Audio-Visual Learning

2026

What Drives Success in Physical Planning with Joint-Embedding Predictive World Models?

2025 · Trans. Mach. Learn. Res.

23 citations

V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning

2025 · arXiv.org

537 citations