Recent years have witnessed new frameworks/algorithms able to deal with multiple views, such as Multiple Kernel Learning (MKL), Boosting, Co-regularized approach. Such algorithms come from the Machine Learning community and find applications in many different areas, such as Multimedia Indexing, Computer Vision, Bio-informatics, Neuro-imaging...
Multiview learning, naturally enough, emphasise the potential benefits of learning through collaboration with multiple sources of data (e.g. video document can be described through images, sound, motion, text). Depending on the context, this issue of learning from multiple descriptions of data goes under the name of multiview learning (machine learning, computer vision), multimodality fusion (multimedia), among others. This workshop is the opportunity to bring together theoretical and applicative communities around multiview learning, which could lead to significant contributions in Machine Learning, Multimedia Mining and Computer Vision.
This workshop builds upon successful previous machine learning workshops on multiview learning or connections between ML and applications, like Machine Learning techniques for processing multimedia content
(ICML 2005), Learning with multiples views
(ICML 2005), Learning from multiples sources
(NIPS 2008), Learning from multiples sources with applications to robotics
(NIPS 2009) where links between theory and applications of the multiview paradigms are made. The literature and the advances on multiview learning have grown up to a point where a broad synthesis is required.
The workshop will be held at Nancy, France, on september 19th.
Topics of Interest
The main objectives of this workshop are to 1) introduce recent development in machine learning with multiview setting, 2) focus on various problematics in multimedia and computer vision where such setting arise and 3) offer new directions and discuss about open questions that appear. In particular, the following topics are relevant:
- Diversity / Complementarity / (Dis-)agreement between the views: How to take benefit of diversity and complementarity between views? How to handle disagreement between views?
- Incomplete / Noisy data: Multiview should help handle noisy data ; is it true? How to manage missing view?
- Missing labels / Noisy annotations: Multiview learning with missing labels or noisy annotations ; Can multiview algorithms be more robust with outliers?
- Large-scale / Big data: Are multiview and big data analysis compatible?
- Ranking / Learn with imbalanced data set: Learning to rank with multiple views ; How to take advantage of multiview setting with imbalanced data set?
- Representation Learning: Bag-of-words and attributes learning with multiple sources ; New deep architecture for multiview learning ; How to learn good representation in multiview setting?
- Domain adaptation / Transfer learning: Is domain adaptation easier with multimodal data?
Format and Publicity
We invite participants to submit papers with contribution in multiview learning that bridge the gap between theoretical advances and applications from Multimedia Mining or Computer Vision. We especially encourage submissions that directly address any of the topics of interest, related papers are also welcomed. The selected submission may be accepted either as an oral presentation or as a poster presentation. Accepted papers will be made available online at the workshop website
Workshop registration will be handled by the main conference, please check ECML/PKDD'14
for more details.