Description: This module introduces the theoretical and methodological foundations of deep learning in the context of ecological forecasting. Students gain an understanding of how deep learning models function, how they can be evaluated, and when they are appropriate for specic ecological forecasting tasks. The module covers model architectures, training principles, performance evaluation, and comparative model assessment. Students learn to critically assess the suitability of different model classes—including convolutional neural networks (CNNs) for spatial data, recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and Transformers for temporal forecasting, as well as graph neural networks (GNNs) for ecological network data. Specialised architectures such as autoencoders, generative adversarial networks (GANs), and diffusion models are also introduced.
Learning outcomes:
Explain the core principles and architectures underpinning deep learning models.
Evaluate the suitability of different deep learning approaches for various ecological forecasting tasks.
Interpret and compare model performance using appropriate validation and benchmarking methods.
Critically discuss the limitations, interpretability, and uncertainty of deep learning models in ecological applications.
Integrate conceptual understanding of deep learning into later thematic and applied modules within the program.
Module group: Foundations of Ecological Forecasting
Credits: 5 ECTS
Lecturers: Prof. Dr. Andreas Hotho