Primary supervisorChristoph Bergmeir
In modern "Big Data" environments often big quantities of related time series are available such as sales time series across different stores and products, measurements from many similar machines e.g. wind farms server farms etc. Forecasting in this case with traditional univariate forecasting procedures leaves great untapped potential for producing more accurate forecasts. Consequently big tech companies such as Amazon, Microsoft, and Uber have started recently to untap the enormous potentials of such datasets using deep learning methods in very successful ways for forecasting on their vast datasets. In particular recurrent neural networks are promising and my research team has already developed various models in this space.
As this area is my main area of research, there are various research topics available, around probabilistic forecasting (researching new loss functions), feature selection/engineering, transfer learning, causal inference, concept drift and online/stream learning, ensembling, local interpretability, evaluation procedures, seasonal factor decomposition, and others. You would be part of a very active research group with several PhD students, external collaborators, and connections to industry. I have topics available that range from rather theoretic to direct applications on real-world data from industry partners.
Hewamalage, H., Bergmeir, C. and Bandara, K., 2019. Recurrent neural networks for time series forecasting: Current status and future directions. arXiv preprint arXiv:1909.00590.
Bandara, K., Bergmeir, C. and Hewamalage, H., 2019. LSTM-MSNet: Leveraging Forecasts on Sets of Related Time Series with Multiple Seasonal Patterns. arXiv preprint arXiv:1909.04293.
Bandara, H, C Bergmeir, and S Smyl (2019). Forecasting Across Time Series Databases using Recurrent Neural Networks on Groups of Similar Series: A Clustering Approach. Expert Systems with Applications. (in press, available online 26 August 2019) https://arxiv.org/pdf/1710.03222.pdf
Bandara, H, P Shi, C Bergmeir, H Hewamalage, Q Tran, and B Seaman (2019). Sales Demand Forecast in E-commerce using a Long Short-Term Memory Neural Network Methodology. In: 26th International Conference on Neural Information Processing (ICONIP) 2019, December 12 - 15, Sydney (Australia).
Python, TensorFlow, Data Science, Machine Learning.
Knowledge in traditional time series forecasting is a plus.