🚇 Sustainability of Hamburg’s U5 project

U5 is comparable in investment scale to Stuttgart 21, running almost entirely underground, which means significant construction-phase greenhouse gas emissions. Hamburger Hochbahn AG’s own life cycle assessment puts emissions at 2.7 million tCO₂e using conventional construction methods.

In his MSc thesis, our student Georg Benjamin Fiedler extended this analysis with economic input-output data to capture upstream emissions outside the LCA’s system boundary. Our upper-bound estimate for conventional construction: 3.6 million tCO₂e.

Furthermore, the transport authorities project up to 70% emission savings during construction, largely contingent on future climate-friendly cement production. Using stochastic sensitivity analysis, we showed that small uncertainties in cement production emissions swing the project’s overall climate balance substantially. Our recommendation: a methodologically robust forecast of cement decarbonisation pathways is needed to substantiate the 70% reduction claim.

Great work by Georg and Sven Lundie, who co-supervised this thesis together with Johannes.

📝 NeurIPS 2026

NeurIPS has long been one of the leading venues for machine learning and computational neuroscience, and it’s an honor to contribute to the review process at this level.

Amid ongoing discussions about review quality, reviewer workload, and the role AI should—or should not—play in peer review, I’m looking forward to doing my part to support rigorous, constructive feedback and impactful research.

Let’s keep discussing the issue of peer review: is it entirely broken or doing better than ever?

Team at GPSD

Our team has been showcasing our work at various workshops and conferences these months. Most recently, Ali, Francesco, Gitte and Mahsa gave talks and presented posters at the 17th German Probability and Statistics Days in Dresden. For example, Ali gave a presentation about geometry-inspired insights into deep-learning architectures, and Mahsa talked about stationary points in deep learning. Well done everyone—let’s go, Hamburg data science! 🧑‍🏫️