Event Details

**Free tutorial, please register as places are limited (*for academics, students and software vendors please see below)**

This tutorial on Forecasting with Artificial Intelligence - from shallow to deep neural networks is aimed at industry practitioners in the area of demand planning and predictive analytics to gain first hands-on experience in how to forecast with Artificial Intelligence using neural networks. You will learn how to apply simple neural networks to various forecasting tasks and understand the benefit of applying deep learning methods, and provided in 2 sections that build upon each other in order to provide a comprehensive insight.

Programme

Tutorial on the Fundamentals of (shallow feedforward) Neural Networks for Forecasting

Sven F. Crone, Assistant Professor at Lancaster University Management School

Content:

Neural networks, the algorithms driving the latest success of artificial intelligence in speech, image and voice recognition, have recently moved to the top of the Gartner hype cycle. Behind their current industry grade successes in real-world case studies is a proud history going back to the foundations of computing and neuroscience in the 1940s, and almost 80 years of successful applications in the industry from parking trucks, recognising sonar signals in the cold war, to Optical Character Recognition and adaptive intelligent control in cars. This tutorial will introduce the fundamentals of artificial neural networks, starting from the biological motivation to early design paradigms, architectures and learning algorithms, to methodologies and live demos in specifying the popular multilayer perceptron neural networks for forecasting using real-time series data found in industry.

Tutorial on Deep and Deep Recurrent Neural Networks

Ralp Grothmann, Principal Expert for Predictive Analytics at Siemens

Content:

The tutorial deals with a new hype wave in neural network modelling called Deep Learning or Deep Neural Networks, building upon the understanding of shallow feedforward neural networks in the preceding session. We will look behind the scenes and will explain the differences between “standard” feedforward and deep neural network models. The decay in gradient information over long chains of hidden layers can be avoided by e.g. multiple outputs, information highways or shortcuts, as well as stochastic learning rules. Auto-associators and convolutions enable us to model high-dimensional input spaces by the automated generation of features. In addition to deep feedforward neural networks, we will also deal with time-delay recurrent neural network architectures, where deepness is a natural feature when non-algorithmic learning techniques like error backpropagation through time are used. Simple recurrent neural networks, long-short-term memory networks (LSTM), echo state networks and large recurrent neural networks are popular examples. We will give examples of the application of deep neural networks form recent project work and show the merits of the “new” approach in comparison to non-deep modelling techniques.

Course details

This course requires no pre-requisites in statistics, forecasting or artificial neural networks. You are welcome to bring your own dataset so that models can be constructed using your dataset in a miniature hackathon in sessions in the evening and next morning. Please note that the course does not cover catering.

All participants of the tutorial are invited to also attend the case-study workshop on “Forecasting with Artificial Intelligence in Industry” on the following day, which will cover multiple talks from academic and industry leaders, also free to attend for industry practitioners.