📝 All Publications

Sonia Laguna, Jorge Da Silva Gonçalves, Moritz Vandenhirtz, Alain Ryser, Irene Cannistraci, Julia E. Vogt (2026). Models Designed to Forget: Machine Unlearning via Key Deletion. In NeurIPS 2026.

Joao Monteiro, Louis Béthune, Anastasiia Filippova, Sonia Laguna, David Grangier, Marco Cuturi (2026). KV-Kaizen: Learning Context-Adaptive Cache Compression Choices. Preprint - Arxiv.

Samuel Ruiperez-Campillo*, Michele Angelo Copetti*, Jorge Da Silva Gonçalves*, Sonia Laguna, Thomas Hofmann, Julia E. Vogt (2026). Cyclostationary Phase Conditioning for Medical Time Series Diffusion. Preprint - Arxiv.

Kilian Bänziger*, Sonia Laguna*, Markus Kreft, Robert Jakob, Kevin O'Sullivan, Lasse B. Strand, Julia E. Vogt (2026). ConRAG: Lightweight Inference of Multi-Hop Relations. Preprint - Arxiv.

Sonia Laguna, Joao Monteiro, Marco Cuturi, Pierre Ablin, Eleonora Gualdoni (2026). Cartridges++: KV Cache Compression without Off-Context Derailment. Preprint - Arxiv.

Eleonora Gualdoni, Sonia Laguna, Louis Béthune, Joao Monteiro, Pierre Ablin, Marco Cuturi (2026). DynaMiCS: Fine-tuning LLMs with Performance Constraints using Dynamic Mixtures. Preprint - Arxiv.

Jonas Mirlach*, Sonia Laguna*, Julia E Vogt (2026). Reference-Guided Machine Unlearning. In ICLR 2026 Workshop AIWILD and extended Preprint - ArXiv.

Sonia Laguna*, Andrea Agostini*, Alain Ryser, Samuel Ruiperez-Campillo, Irene Cannistraci, Moritz Vandenhirtz, Stephan Mandt, Nicolas Deperrois, Farhad Nooralahzadeh, Michael Krauthammer, Thomas M Sutter, Julia E. Vogt (2026). Structure is Supervision: Multiview Masked Autoencoders for Radiology. In Transactions in Machine Learning Research (TMLR) 2026.

Wiktor Jan Hoffmann, Sonia Laguna, Moritz Vandenhirtz, Emanuele Palumbo, Julia E. Vogt (2026). Post-hoc Stochastic Concept Bottleneck Models. In ICLR 2026 Workshop Trustworthy AI.

Moritz Vandenhirtz, Samuel Ruiperez-Campillo, Sonia Laguna, Simon Böhi, Irene Cannistraci, Andrea Agostini, Ece Ozkan, Thomas M Sutter, Julia E. Vogt (2026). Foundation Model for Cardiac Time Series via Masked Latent Attention. In ICLR 2026 Workshop FM4Science.