Time series data is essential in fields like finance, energy, healthcare, and climate science. This advanced course builds on foundational knowledge and equips participants with sophisticated techniques for forecasting and anomaly detection.
The course begins with deep learning for time series, starting with multilayer perceptrons (MLPs) before progressing to Variational Autoencoders (VAEs) and diffusion models. We also cover probabilistic approaches including Gaussian processes for principled uncertainty estimation.
Additionally, the course explores advanced anomaly detection methods, including VAE-based approaches and Matrix Profile for motif and discord discovery. Ensemble methods for combining forecasts and improving robustness are also covered.
Through hands-on exercises and real-world examples, participants will develop practical skills for applying advanced time series methods.
This course is ideal for those with foundational knowledge who want to expand their toolkit. Completion of Machine Learning for Time Series: Introduction or equivalent knowledge is recommended.
Learning objectives
- Understand Deep Learning Architectures for Time Series.
- Apply Probabilistic Approaches and Uncertainty Quantification.
- Implement Advanced Anomaly Detection Methods.
- Combine Models Using Ensemble Methods.
- Apply Advanced Methods to Complex Real-World Problems.
