x-dl4ts
Deep Learning for Time Series
Welcome to the course materials for Deep Learning for Time Series. This course is part of the data science master programme at IP Paris. The course provides a comprehensive dive into modern neural network architectures specifically tailored for temporal data, ranging from fundamental concepts to advanced sequence modeling.
- Target Audience: Data scientists, researchers, and students with a basic knowledge of Machine Learning and Deep Learning.
- Github Repository for the course materials (Notebooks for lab sessions, Typst code for slides, etc.)
Course Modules
Introduction to DL for Time Series
 |
Overview: A quick overview of “standard” (ie. non-deep) approaches for time series forecasting. Slide Deck (PDF) |
Practical aspects of forecasting
 |
Overview: From multi-step forecasting to uncertainty estimation: practical tips and tricks. Slide Deck (PDF) |
Convolutional Neural Networks & Recurrent Neural Networks
 |
Overview: Extracting local patterns vs capturing temporal dependencies. Slide Deck (PDF) |
Attention Mechanisms & Transformers
 |
Overview: How self-attention revolutionized sequence modeling. Slide Deck (PDF) |
Time Series Classification
 |
Overview: How to classify sequences using deep learning architectures. Slide Deck (PDF) |
Time Series Foundation Models
 |
Overview: Large pre-trained models and transfer learning. Slide Deck (PDF) |
Continuous-Time Models
 |
Overview: Modeling irregularly sampled data and continuous-time dynamics. Slide Deck (PDF) |
State-Space Models
 |
Overview: Combining deep learning with state-space modeling for time series analysis. Slide Deck (PDF) |
About the Instructor
Romain Tavenard is a Professor at Université de Rennes 2 and a researcher specializing in Machine Learning for Time Series. He is the creator of the tslearn library.