Machine Learning for Temporal and Dynamical Data

Topics: Temporal Data, Event Data, Forecasting, Temporal Point Processes

Modeling temporal data is one of the fundamental tasks in machine learning since many systems keep track of signals that are changing over time. One example is using historical power consuption to forecast the future consumption - a crucial task for smart grids. Beyond forecasting, we may use machine learning methods to find anomalies in sensor data which can provide huge industrial value. These methods are also the stepping stone to autonomous driving, where a machine learning model should recognize and adapt to changing road situation, from sensor data gathered in real time.

An even more challenging setting is when we are dealing with asynchronous event data or time series that are sampled irregularly. For example, such event streams are common in financial systems or electronic health records. The framework of Temporal Point Processes provides the theoretical foundations for dealing with such continuous-time event data. However, a lot of challenges remain open when it comes to applying such models in practice:

  • How do we create models that accurately predict the future?
  • How can we understand the temporal dynamics of the system?
  • How can we quantify uncertainty in our predictions?

SELECTED PUBLICATIONS

  • Marin Biloš, Johanna Sommer, Syama Sundar Rangapuram, Tim Januschowski, Stephan Günnemann
    Neural Flows: Efficient Alternative to Neural ODEs
    Neural Information Processing Systems (NeurIPS), 2021
  • Oleksandr Shchur, Ali Caner Turkmen, Tim Januschowski, Jan Gasthaus, Stephan Günnemann
    Detecting Anomalous Event Sequences with Temporal Point Processes
    Neural Information Processing Systems (NeurIPS), 2021
  • Oleksandr Shchur, Ali Caner Türkmen, Tim Januschowski, Stephan Günnemann
    Neural Temporal Point Processes: A Review
    International Joint Conference on Artificial Intelligence (IJCAI), 2021
  • Oleksandr Shchur, Nicholas Gao, Marin Biloš, Stephan Günnemann
    Fast and Flexible Temporal Point Processes with Triangular Maps
    Neural Information Processing Systems (NeurIPS), 2020
  • Richard Kurle, Syama Sundar Rangapuram, Emmanuel de Bézenac, Stephan Günnemann, Jan Gasthaus
    Deep Rao-Blackwellised Particle Filters for Time Series Forecasting
    Neural Information Processing Systems (NeurIPS), 2020
  • Oleksandr Shchur, Marin Biloš, Stephan Günnemann
    Intensity-Free Learning of Temporal Point Processes
    International Conference on Learning Representations (ICLR), 2020
  • Richard Kurle, Botond Cseke, Alexej Klushyn, Patrick van der Smagt, Stephan Günnemann
    Continual Learning with Bayesian Neural Networks for Non-Stationary Data
    International Conference on Learning Representations (ICLR), 2020
  • Marin Biloš, Bertrand Charpentier, Stephan Günnemann
    Uncertainty on Asynchronous Time Event Prediction
    Neural Information Processing Systems (NeurIPS), 2019
  • Artur Mrowca, Martin Nocker, Sebastian Steinhorst, Stephan Günnemann
    Learning Temporal Specifications from Imperfect Traces Using Bayesian Inference
    Design Automation Conference (DAC), 2019