Multi-Task Learning: a Bayesian approach
Event details
| Date | 18.12.2012 |
| Hour | 13:00 › 14:00 |
| Speaker | Dr. Cedric Archambeau, Machine Learning for Services group at Xerox Research Center Europe |
| Location | |
| Category | Conferences - Seminars |
In many real life problems multiple related target variables need to be predicted from a single set of input features. A problem that attracted considerable interest in recent years is to label an image with (text) keywords based on the features extracted from that image. In general, this multi-label classification problem is challenging as the number of classes is equal to the vocabulary size and thus typically very large. While capturing correlations between the labels seems appealing it is in practice difficult as it rapidly leads to numerical problems when estimating the correlations. In this talk, I will introduce a sparse Bayesian model for multi-task regression and classification. The model is able to capture correlations between tasks, while being sparse in the features. This is especially attractive for the interpretation of the results. By adopting a Bayesian approach we can learn the level of sparsity from the data. Empirical evaluations on data sets from biology and vision demonstrate the applicability of the model, where on both regression and classification tasks it achieves competitive predictive performance compared to previously proposed methods. I will also briefly discuss how the model was applied to the analysis of data collected in Xerox call centres.
This is joint work with Shengbo Guo and Onno Zoeter.
This is joint work with Shengbo Guo and Onno Zoeter.
Practical information
- Informed public
- Free
- This event is internal
Organizer
- LAPMAL Seminar
Contact
- Prof. Matthias Seeger ([email protected])