Time Series Predictions with Sequential Neural Networks (infBSemZsnN-01a)
Abstract
​
This seminar addresses the learning of models for the analysis and prediction of time series. Time series are temporarily ordered data points that are collected e.g. hourly, weekly or annually. The analysis and prediction of time series are of great interest for a variety of different fields (e.g. economics, remote sensing, robotics, meteorology). The aim of time series models is to enable the prediction of trends that make it possible to derive statements on the future development of observations.
Learning Objectives
​
The students will learn to ...
​
-
Understand techniques for processing temporally organized data
-
Know different architectures of sequential neural networks
-
Are able to implement and train sequential neural networks
Course Content
​
The aim of the seminar is to learn models for temporally organized data with the help of sequential neural networks. In the seminar, different architectures of neural networks are presented (LSTM, ConvNets, Transformer) and their possibilities for analyzing and predicting time series are examined. The course aims in particular to acquire skills in dealing with real data and the implementation of sequential neural network architectures. Particularly with regard to practical applications, students should acquire skills that enable them to train and use their own models for time series prediction. As part of the seminar, students will work on a time series analysis problem alone or in groups of two with the aim of developing and training their own sequential model.
Further Requirements
​
​Knowledge of Python programming.
Teaching and Learning Methods
Learning materials will be provided in the form of presentation slides. Primary lecture media is projected slide presentation. Occasionally complemented with drafts on board/white board. Concepts are introduced in the lectures with the help of examples and specific application tasks. In the exercise the knowledge is deepened and applied - guided by weekly homework assignments.
Literature
-
Deep Learning, Ian Goodfellow and Yoshua Bengio and Aaron Courville, 2016, MIT Press