Prof. Pascal Frossard leads the LTS4 research laboratory at EPFL (Lausanne, Switzerland), internationally recognised for its contributions to machine learning, graph representation learning, and AI for science. The LTS4 lab pioneered the field of Graph Signal Processing (GSP), developing foundational theory, algorithms, and applications for data defined on irregular structures such as graphs and networks. These contributions include early work on sparse representations, graph learning, and equivariant models for network data, precursors to modern graph neural network (GNN) architectures.

​

A cornerstone of the lab’s recent research is generative modelling for graphs. In particular, the DiGress framework established discrete diffusion as a state-of-the-art methodology for generating graphs with categorical node and edge attributes. Subsequent work has extended these models to high-impact applications, including drug discovery, digital pathology (e.g. generation of tertiary lymphoid structure cell-graphs), and biological tissue reassembly. The lab also addresses key challenges posed by large-scale AI and foundation models, with a focus on robustness, generalisation, interpretability, and efficient strategies for model merging and fine-tuning.

​

As co-director of the Swiss National AI Institute (SNAI) and Academic Director of the EPFL AI Center, Prof. Frossard plays a key role in keeping the SATURNA program at the forefront of AI technologies, while ensuring strong alignment with and visibility within the Swiss AI research landscape.