PhD Position: Geometry and physics-based inductive bias for robust and explainable foundational[...]

PhD Position: Geometry and physics-based inductive bias for robust and explainable foundational models – AI4I Deadline December 17th 2025 at 12:00 (noon – CET) Curriculum Hostile and Unstructured Environments (CODE 11660), Topic number 2 Hosting Institution AI4I – The Italian Institute of Artificial Intelligence Funding Scheme This doctorate grant is fully funded by The Italian Institute of Artificial Intelligence (AI4I) in collaboration with Università degli Studi di Genova Position Description The Italian Institute of Artificial Intelligence (AI4I), in collaboration with Università degli Studi di Genova, invites applications for a PhD position in Geometry and physics-based inductive bias as part of the PhD Programme in Hostile and Unstructured Environments. Recent research highlights the promising potential of geometry‑aware and physics‑informed attention mechanisms, and the power of differential geometry in the analysis and refinement of foundational models. This PhD will study the properties of Riemannian embeddings in large multimodal models, exploring how different curvatures can accommodate and associate inter‑connected tokens. We will design geometry‑, physics‑, and control‑based attention mechanisms to capture complex perception‑action interactions, potentially leading to new models that leverage physics‑aware, curved spaces to better abstract and disentangle information from multimodal perception systems. The developed methods will be applied to general foundational models, vision‑action large models for robotics, and related areas. Responsibilities • Conduct research on Riemannian embeddings and curvature‑based attention mechanisms. • Develop and evaluate geometry‑ and physics‑based models for multimodal foundational systems. • Publish findings in peer‑reviewed journals and present at international conferences. • Collaborate with interdisciplinary research teams and industry partners. • Contribute to the dissemination of results through workshops, tutorials, and open‑source releases. Requirements Must have skills: Excellent Master’s degree in computer science, physics, mathematics, electrical or mechatronics engineering, or a related field Strong background in machine learning and robotics Good programming skills in Python Fluent in spoken and written English Team player, but able to work autonomously Experience with scientific writing Good to have skills: Background in applied differential geometry Publication of peer‑reviewed research papers References A. Testa et al. “Geometric Contact Flows: Contactomorphisms for Dynamics and Control”, ICML, 2025. N. He, et al. “Beyond Euclidean: Foundation Models Should Embrace Non‑Euclidean Geometries”, WWW, 2025. H. Ding et al. “Fast and Robust Visuomotor Riemannian Flow Matching Policy”, Transactions on Robotics, 2025. Number of Positions Available 1 Application Documents One‑page cover letter including a short (two‑paragraph) research proposal related to the PhD topic and aligned with your professional interests. Bachelor’s and Master’s diplomas with transcripts and grades. Main Research Site AI4I – The Italian Institute of Artificial IntelligenceCorso Castelfidardo 22, 10129 Torino, Italy What We Offer Access to high‑performance computing resources and advanced research infrastructure. Opportunities for international collaboration and contributions to high‑impact publications. A dynamic and interdisciplinary research environment. Financial support for attending international conferences and Winter/Summer schools. Start Date 1st March 2026 How to Apply Applications for this position are managed by Università degli Studi di Genova. Please apply via the university’s official PhD admissions portal. #J-18808-Ljbffr