Why Do Multi-Stream, Multi-Band and Multi-Modal Approaches Work on Biometric User Authentication Tasks?. Poh, N. & Bengio, S. In IEEE International Conference on Acoustic, Speech, and Signal Processing, ICASSP, volume 5, pages 893–896, 2004.
Why Do Multi-Stream, Multi-Band and Multi-Modal Approaches Work on Biometric User Authentication Tasks? [link]Paper  abstract   bibtex   
Multi-band, multi-stream and multi-modal approaches have proven to be very successful both in experiments and in real-life applications, among which speech recognition and biometric authentication are of particular interest here. However, there is a lack of a theoretical study to justify why and how they work, when one combines the streams at the feature or classifier score levels. In this paper, we attempt to cast a light onto the latter subject. While there exists literature discussing this aspect, a study on the relationship between correlation, variance reduction and Equal Error Rate (often used in biometric authentication) has not been treated theoretically as done here, using the mean operator. Our findings suggest that combining several experts using the mean operator, Multi-Layer-Perceptrons and Support Vector Machines \emphalways perform better than the \emphaverage performance of the underlying experts. Furthermore, in practice, \emphmost combined experts using the methods mentioned above perform better than \emphthe best underlying expert.
@inproceedings{poh:2004:icassp,
  author =   {N. Poh and S. Bengio},
  title =    {Why Do Multi-Stream, Multi-Band and Multi-Modal Approaches Work on Biometric User Authentication Tasks?},
  booktitle = {{IEEE} International Conference on Acoustic, Speech, and Signal Processing, {ICASSP}},
  year =   2004,
  volume = 5,
  pages = {893--896},
  url = {publications/ps/poh_2004_icassp.ps.gz},
  pdf = {publications/pdf/poh_2004_icassp.pdf},
  djvu = {publications/djvu/poh_2004_icassp.djvu},
  original = {2004/multi_stream_authentication_icassp},
  idiap = {publications/pdf/rr03-59.pdf},
  topics = {multimodal,biometric_authentication},
  web = {http://dx.doi.org/10.1109/ICASSP.2004.1327255},
  abstract = {Multi-band, multi-stream and multi-modal approaches have proven to be very successful both in experiments and in real-life applications, among which speech recognition and biometric authentication are of particular interest here. However, there is a lack of a theoretical study to justify why and how they work, when one combines the streams at the feature or classifier score levels.  In this paper, we attempt to cast a light onto the latter subject.  While there exists literature discussing this aspect, a study on the relationship between correlation, variance reduction and Equal Error Rate (often used in biometric authentication) has not been treated theoretically as done here, using the mean operator. Our findings suggest that combining several experts using the mean operator, Multi-Layer-Perceptrons and Support Vector Machines \emph{always} perform better than the \emph{average performance} of the underlying experts. Furthermore, in practice, \emph{most} combined experts using the methods mentioned above perform better than \emph{the best underlying expert}.},
  categorie = {C},
}

Downloads: 0