Can We Trust Computers That See? Computer Vision and Earth System Sciences.

Can We Trust Computers That See? Opportunities and Risks of Modern AI
The talk highlights the rapid development of artificial intelligence (AI) in the field of machine vision, using current research projects and practical application areas such as autonomous driving, medical diagnostics (e.g., skin cancer detection), and biodiversity monitoring through camera traps. A central focus is on the challenges posed by model bias, which can arise from the training data, the lack of explainability of complex models, and the absence of causal mechanisms. The talk advocates for a critical approach to AI systems and outlines how integrating world knowledge, explainable AI, and causal inference can lead to the development and application of more robust and trustworthy models.
Joachim Denzler is a full Professor of Computer Vision at the University of Jena. He has been the founding Director of the Michael-Stifel-Center Jena for Data-driven and Simulation Science and the ELLIS Unit Jena (www.ellis-jena.ai) as well as the Director of the Institute of Data Science of the German Aerospace Center (DLR). Joachim Denzler’s main research interests revolve around the analysis, prediction and understanding of complex dynamical systems, including applications from medicine, psychology and earth system sciences. Fine-grained object classification, active learning and causal inference for time-series analysis are of particular interest. He addresses these topics with the development and application of machine learning methods, including deep learning, and aspects from explainable AI. Joachim Denzler is a member of the board of the Thuringia Center for Learning Systems and Robotics (www.tzlr.de) with the mission to transfer research results from AI to industry. Joachim is excited about the potential of using applications as drivers for basic research, especially to contribute to our society’s urgent and pressing problems, like climate change and biodiversity loss.
Joachim Denzler published more than 500 papers at international conferences and journals with around 10000 citations and an h-index of 52, according to google scholar. He is a PC member and reviewer of major conferences (NeurIPS, ICCV, ECCV, CVPR, ICLR) and Journals (IEEE TPAMI, IJCV, etc.). His group consists of 15 PhD students and receives funding from the German Research Foundation (DFG), Federal Ministry of Science (BMBF), and EU, as well as from industrial projects. He is a member of IEEE and IEEE Computer Society.


Learning Earth systems with AI: From observations to prediction
Artificial intelligence is becoming a key tool in understanding and predicting how Earth systems are changing. From climate variability to water and carbon cycles, many environmental processes are difficult to observe directly, vary across space and time, and depend on complex physical interactions. This talk will discuss how AI methods can help extract relevant information from diverse data sources, improve predictive accuracy under uncertainty, and support scientific reasoning. It will focus in particular on approaches that combine machine learning with physical models, showing how these can complement traditional simulations and observational techniques. Several problem types will be outlined to illustrate the practical use of AI in this context. The talk will also consider ongoing challenges, including generalization, data gaps, and model interpretability, and discuss how similar methods may be applied in other domains.
Dr. Shijie Jiang is the leader of the ELLIS Unit Jena Research Group "Machine Learning for Hydrological and Earth Systems (ML4HES)" at the Max Planck Institute for Biogeochemistry in Jena. Since August 2023, he has been heading this group, which is funded by the Carl Zeiss Foundation. His research focuses on understanding the interactions between climate, water, and ecosystems by integrating data and domain knowledge using hybrid and explainable machine learning methods. His current work explores terrestrial ecohydrological processes and interactions across different scales, the integration of physics and data through hybrid and explainable AI, coupling and feedback mechanisms between the water, energy, and carbon cycles, the predictability, attribution, and impacts of climate extremes, as well as hydroclimatology, climate change impacts, and human influences. Shijie Jiang earned his Ph.D. from the National University of Singapore in 2021, where he focused on overcoming challenges in monitoring, modeling, and understanding hydrological systems using AI approaches. During his postdoctoral research at the Helmholtz Centre for Environmental Research, he worked on data-driven process understanding of extreme climate events. His work has received wide recognition, with several publications ranked among the most cited and most downloaded in their respective journals. Since 2022, he has also served as a convener for sessions on (explainable) machine learning in Earth sciences at major international conferences such as EGU and AGU.