Computational Applied Mathematics & AI Lab

ELLIS Institute Tübingen and Max Planck Institute for Intelligent Systems

tuebingen_pic.jpg

The Computational Applied Mathematics & AI Lab (CAMAIL) is a research group at the ELLIS Institute Tübingen and the Max Planck Institute for Intelligent Systems headed by T. Konstantin Rusch.

The research of CAMAIL lies at the intersection of AI and Computational Applied Mathematics. A central focus is the design of novel methods that improve the efficiency, capabilities, and theoretical understanding of AI systems, including efficient architectures and compression techniques such as pruning and quantization. We also investigate reasoning in emerging AI paradigms, including looped models and diffusion-based language models. In parallel, we study how AI can advance fundamental algorithms in computational mathematics. These methods and insights are subsequently applied to challenges in physics, computational science, and robotics.

We are hiring. If you are interested in joining our group as a PhD student, postdoc or intern, feel free to reach out via email. Please include a CV and transcripts of your grades when first reaching out. We will admit PhD students through the ELLIS, CLS, and IMPRS programs, the application deadlines are usually around November.

news

Sep 24, 2026 Papers accepted at NeurIPS 2026: Our lab will present 3 full papers at NeurIPS 2026, on stable and adaptive deep looped transformers (fixed-point reasoners), calibration-free bit allocation for MoE quantization (AlphaQ), and a benchmark for scientific time series (PhyTS).
Apr 20, 2026 Papers accepted at ICML 2026: Our lab will present 3 papers at ICML 2026 (1 full paper, and 2 workshop papers), on neural low-discrepancy sequences (NeuroLDS), stable and adaptive deep looped transformers (fixed-point reasoners), and state reduction in linear attention.
Jan 26, 2026 Papers accepted at ICLR 2026: Our lab will present 4 papers at ICLR 2026 in Rio (1 full paper, and 3 workshop papers), on control-theoretic in-training compression of SSMs, frequency-aware flow matching, structured pruning of (gated) deltanet models, and data-free mixed-precision quantization of MoEs.