Linear video coding
Conventional and learning-based video compression — codecs, rate–distortion, and the boundary between hand-designed and neural pipelines.
Active research group · LTCI
Multimedia team - at LTCI, Télécom Paris, Institut Polytechnique de Paris
The Multimedia (MM) team conducts research on the full life-cycle of multimedia data: from compression and transport to learning, explainability, and generation. We work where signal processing meets modern deep learning — with an emphasis on frugal, geometric, and multimodal approaches.
128+
Publications
2021–2026, peer-reviewed
29
Team members
faculty, postdocs & students
7
Research themes
from coding to learning
A*
Top venues
CVPR · ICCV · NeurIPS · ICLR
CVPRICCVNeurIPSICLRAAAIWACVECCVICMLACM MultimediaIEEE TPAMIIEEE TMMIEEE TIPIEEE TCSVTPattern Recognition
CVPRICCVNeurIPSICLRAAAIWACVECCVICMLACM MultimediaIEEE TPAMIIEEE TMMIEEE TIPIEEE TCSVTPattern Recognition
What we work on
All themes →Rich-media scene representation
Representing complex scenes with multiple media, layouts, and interactions — beyond a single video stream.
Multimedia content adaptation
Adapting multimedia content to networks, devices, and users — quality of experience under constraints.
Multimedia distribution networks
Transport, orchestration, and protocols for delivering multimedia at scale.
Frugal and efficient AI
Pruning, quantization, low-rank methods, and other tools to make deep models small enough to deploy.
Geometric deep learning
Learning on graphs, manifolds, and structured domains — where the geometry of the data shapes the architecture.
Latest news
All news →New ANR JCJC project
Many congratulations to Stephan Alaniz who has received funding for his ANR JCJC project (VieLM).
- ECCV 2026
Three papers accepted at ECCV 2026
The MM team has three accepted contributions at ECCV 2026.
- CVPR 2026
Three papers accepted at CVPR 2026
The MM team has three accepted contributions at CVPR 2026, spanning compression, geometric deep learning, and multimodal learning.
Recent work
All publications →AracNet: Revealing Debiasing Signals across Layers with Shallow Monitors
Look But Don't Touch with Sparse Autoencoders for Unlearning in Diffusion Models
Training-free Uncertainty Guidance for Complex Visual Tasks with MLLMs
From Deep to Shallow: Unconstrained and Efficient Layer Merging Strategy
THE SILENCE OF THE WEIGHTS: AN INVESTIGATION OF STRUCTURAL PRUNING STRATEGIES FOR ATTENTION-BASED AUDIO SIGNAL ARCHITECTURES
CutClean: Neural Network Pruning for Privacy-Preserving Inference