Journal Club

In this talk, we discuss the paper "Deep learning for universal linear embeddings of nonlinear dynamics" by B. Lusch, J. N. Kutz, and S. Brunton, Nat. Comm. 2018. Abstract  Identifying coordinate transformations that make strongly nonlinear dynamics approximately linear has the potential to enable nonlinear prediction, estimation, and control using linear theory. The Koopman operator …

In this talk, we discuss the paper "Large language models for scientific discovery in molecular property prediction" by Yizhen Zheng et.al., nature machine intelligence, 2025. Abstract Large language models (LLMs) are a form of artificial intelligence system encapsulating vast knowledge in the form of natural language. These systems are adept at numerous complex tasks including …

In this talk, we discuss the paper, "Data splitting to avoid information leakage with DataSAIL" by Roman Joeres, et al., Nature Communications, 2025. Abstract Information leakage is an increasingly important topic in machine learning research for biomedical applications. When information leakage happens during a model’s training, it risks memorizing the training data instead of learning …

In this talk, we discuss the paper "Machine learning methods trained on simple models can predict critical transitions in complex natural systems" by  Smita Deb, Sahil Sidheekh, Christopher F. Clements, Narayanan C. Krishnan, and Partha S. Dutta, in Royal Society Open Science, (2022). Abstract:  Forecasting sudden changes in complex systems is a critical but challenging task, with previously developed methods varying widely in their reliability. Here we develop a novel detection method, using simple theoretical models to train a deep neural network to detect critical transitions—the Early Warning Signal Network (EWSNet). We then demonstrate that this network, trained on simulated data, can reliably predict observed real-world transitions in systems ranging from rapid climatic change to the collapse of ecological populations. Importantly, our model appears to capture latent properties in time series missed by previous warning signals approaches, allowing us to not only detect if a transition is approaching, but critically whether the collapse will be catastrophic or non-catastrophic. These novel properties mean EWSNet has the potential to serve as an indicator of transitions across a broad spectrum of complex systems, without requiring information on the structure of the system being monitored. Our work highlights the practicality of deep learning for addressing further questions pertaining to ecosystem collapse and has much broader management implications.