Machine Learning for Time Series
Objectives
When learning from structured data such as time series data, special attention has to be paid to the models used. Indeed, designing machine learning models requires thinking of the invariants to be learned and either encoding them in the model or designing the model so that it is able to discover such invariants and encode them. In this course, we will cover the use of alignment-based methods in traditional machine learning models. Dedicated neural network architectures will also be tackled. All these models will be illustrated on real datasets. After this course, the student will be able to choose an adequate machine learning model and apply it for a given time series task.
Course outline
– Shift invariance in time series
– Alignment-based methods for time series
– Recurrent neural networks
– Convolutional models for time series
Prerequisites
Basics of neural networks