Lezing
Workshop: Metalearning and model management for data science
- Dr. Carlos Soares
- Datum
- woensdag 6 april 2016
- Tijd
- Locatie
-
Snellius
Niels Bohrweg 1
2333 CA Leiden - Zaal
- B3
This workshop is given by visiting scholar dr. Carlos Soares from the University of Porto. The workshop is open to all interested parties, but specifically targeted towards MSc and PhD students. No registration is required.
Abstract
Traditional methodologies are being challenged by changes in the decision support needs of companies as we move into the “data science” era. Among other things, these changes include a much larger number of models to be used (hundreds or even thousands) and the dynamic nature of the phenomena being modeled (often streaming data with changing distributions). Under these conditions, ensuring that the best (or acceptable) models are being used is not possible using traditional DM methodologies. New methodologies are required to manage models, namely their generation, monitoring and selection. One approach that can be used for this purpose is metalearning. Metalearning consists of applying machine learning approaches to obtain models that relate the characteristics of problems with the performance of methods. Despite the potential, the metalearning field is still at an early stage of its development. In this workshop we will discuss the potential of metalearning for model management and the challenges involved.