AI lounge: Real world machine learning

Welcome to this new AI Lounge. We are happy to present two new interesting presentations.

Poster about the AI lounge (pdf).

Coffee, tea and cookies will be served. 

Looking forward to seeing you at the first AI lounge after the summer break!

Real world machine learning I

What are the challenges of applying machine learning to real world problems? What do we do if we cannot accept a black box solution?

by  Signe Riemer-Sørensen - SINTEF

End-to-end learning is a very flexible and powerful approach where data-preprocessing, feature selection and in some cases even method-selection are not considered individual steps, but are dealt with internally in the analysis. However, many real world problems do not have the data quality or quantity for this approach. I will discuss various approaches to hybrid analytics where the machine learning models are combined with domain knowledge to compensate for data quality.

Real world machine learning II

by Volker Hoffmann - SINTEF

Volker Hoffmann is a Research Scientist at SINTEF Digital. After his PhD in Computational Astrophysics (Zurich), Volker worked with large-scale processing of optical satellite imagery, whereafter he joined SINTEF. At SINTEF, his focus is on how linking data from various sources can solve problems using data-driven methods.

While our world is becoming ever more dependent on electricity, its generation, storage, and consumption patterns are becoming more erratic. To ensure reliability of supply, we must expand our capabilities to understand and forecast grid conditions. Fortunately, the Norwegian power grid is monitored by hierarchies of sensor equipment with detailed failure logs available. Using this rich historical data, we can use machine learning to build predictive models capable of forecasting grid conditions. In this presentation, we describe two examples. First, a model to forecasting power quality fluctuations based on the pre-fault signal characteristics. Second, a model to calculate the risk-level for a large-scale outage based on the expected weather conditions.





Andrea Gasparini, Anne Schad Bergsaker and Thomas Röblitz
Tags: AI, machine learning, SINTEF, deep learning, USIT, ITF, UB, AI lounge
Published Aug. 16, 2019 11:37 AM - Last modified Aug. 27, 2019 12:17 PM