Multi-way arrays

Using data-driven design methodologies in the development of prediction models, for example video quality prediction models, the data is usually  represented by a two-way array or matrix. In many applications, however, the data does have a multi-way nature and in flattening the data into a two-way representation, valuable information is lost in the model building process.

A better alternative is therefore the use of multi-way data analysis methods, for example multi-way PLSR, that consider all directions of the data in the process. In my contributions, I applied this approach so far to the design of video quality metrics, avoiding the temporal pooling usually applied,  resulting in an overall better prediction performance of the metrics.

Key publications:

Video is a Cube, Signal Processing Magazine, IEEE, vol. 28, no. 6, pp. 41-49, 2011. ISSN 1053-5888.


Design of no-reference video quality metrics with multiway partial least squares regression, Quality of Multimedia Experience (QoMEX), 2011 Third International Workshop on, pp. 49-54, 2011. ISBN 978-1-4577-1334-7.


Hybrid No-Reference Video Quality Metric Based on Multiway PLSR, European Signal Processing Conference (EUSIPCO),2012, pp. 1244-1248, 2012. ISBN 978-1-4673-1068-0. ISSN 2219-5491.


Christian Keimel
Christian Keimel
Artificial Intelligence Research Project Leader

My research interests include AI/Machine Learning in the context of Visual Information Processing.