Recurrent Neural Networks for Forecasting

Talk given at

• Amazon, Berlin, June 2019.
• Uber Technologies, San Francisco, September 2019
• Walmart Labs, San Bruno, September 2019
• Monash University, October 2019
• Wesfarmers Analytics Conference, Melbourne, November 2019

Abstract

In this seminar, I’ll talk about recurrent neural networks (RNN) for time series forecasting. I’ll present a comprehensive study around performance of different network architectures, preprocessing procedures, etc on different datasets, and a study around how to best address multiple seasonalities when using RNNs.

Papers

• Hansika Hewamalage, Christoph Bergmeir, Kasun Bandara (2019) Recurrent Neural Networks for Time Series Forecasting: Current Status and Future Directions. In: Working Paper

• Kasun Bandara, Christoph Bergmeir, Hansika Hewamalage (2020) LSTM-MSNet: Leveraging Forecasts on Sets of Related Time Series with Multiple Seasonal Patterns. In: IEEE Transactions on Neural Networks and Learning Systems