Our paper “A Multiscale Interpretability Framework for Identifying Actionable Road Network Features to Mitigate Congestion in Highly Congested Cities” has been published in Scientific Reports!

The paper, led by Nishant Kumar together with Yatao Zhang, Nina Wiedemann, Jimi Oke and Martin Raubal, presents a multiscale framework for identifying road network features associated with traffic congestion across different spatial resolutions. By systematically analysing seven highly congested cities worldwide, we show that the importance of road network features varies considerably between cities and across spatial scales. While infrastructure-related patterns require city-specific planning approaches, several features associated with increased congestion show consistent patterns across cities, offering potential insights for demand-side interventions.

The study offers practical guidance for researchers and urban planners seeking to develop more robust, context-sensitive congestion mitigation strategies and design congestion-resilient cities.

Check out the paper in Scientific Reports!

We welcome Juliette Sikking who will start her PhD at the MIE Lab on the topic of GeoAI and spatial machine learning for human mobility and geospatial data. Juliette is interested in the interplay between machine learning methods and geospatial patterns in urban systems. She is excited about applying spatial machine learning to improve our understanding of human mobility. Before joining ETH, she received her master’s degree in Mathematics from EPFL, with a minor in Quantum Science & Engineering, where her research experience spanned geospatial machine learning, large language models, quantum information theory, and quantum machine learning. She received her bachelor’s degree from EPFL in Mathematics.

Our paper “UrbanFusion: Stochastic Multimodal Fusion for Contrastive Learning of Robust Spatial Representations” was accepted to the International Conference on Machine Learning (ICML 2026)!

In the paper that was led by Dominik Mühlematter as his master thesis at the MIE lab, we are presenting a new method to learn general embeddings of spatial locations that can be used for a variety of downstream tasks. We propose to train with a contrastive loss on masked modality subsets, combined with a reconstruction loss. This scheme not only results in improved performance on downstream tasks such as inferring land use, health indicators or sociodemographics, but is also theoretically shown to retain more information than baseline methods.

The paper will be presented at the ICML conference in Seoul, South Korea, July 6th-11th.

Paper: https://arxiv.org/abs/2510.13774
Code: https://github.com/DominikM198/UrbanFusion

We welcome Wolin Jiang who will start his PhD at the MIE Lab on the topic of causal inference for car sharing interventions. Wolin is interested in the interplay between individual travel behavior and mobility stakeholders in urban systems. He is excited about applying causal inference to improve shared mobility. Before joining ETH, he received his master’s degree from UC Berkeley, where he focused on using reinforcement learning and game theory to optimize on-demand shared ride operations. He received his bachelor degree in Civil Engineering at Tongji University, Shanghai.

Nina Wiedemann, a postdoctoral researcher at the MIE Lab, has been awarded the Förderpreis Geoinformatik by Runder Tisch GIS e.V. for her dissertation titled “Spatio-temporal effects in GeoAI: from predictability to evaluation.”
She presented her work at the Münchner GI Runde on March 19, where she was selected as the award recipient by an independent jury.