Panel: The Influence of Internet Hypes on Multimedia Information Retrieval Research

The research in the multimedia information retrieval (MIR) field started in late 1980ies by addressing the general problem of finding images that fit the query image in terms of color composition and the topology of image regions. Through the rapid development of theory and algorithms for multimedia content analysis in the 1990ies, these efforts soon expanded to address search and retrieval challenges in video and audio collections as well. While these initial efforts can be characterized as largely domain-agnostic and fundamental in nature, this characterization holds less for the next generation of MIR research approaches. These approaches have largely been influenced by the hype around the TRECVID evaluation benchmark and have addressed the challenge of semantic-concept based multimedia indexing and search. Not only that these research approaches have shifted from fundamental issues towards rather straightforward and often too simplistic applications of proven machine learning solutions, but they have also increasingly become biased towards specific sets of labels (ontologies) and content categories. This trend of enlarging the domain bias and abandoning the quest for solutions to truly fundamental MIR problems has even been amplified in the recent years through the hypes around new emerging applications, services and platforms for generating, searching, sharing, indexing and managing multimedia content on the Internet, typical examples of which are Twitter, Flickr, Google and YouTube.

This panel will focus on the trend described above and discuss its consequences regarding the nature, relevance and long-term impact of the research conducted in the MIR community. In particular, we will address the following questions:

1. What are the benefits of actively following Internet hypes when developing the MIR theory and algorithms?
2. What are the consequences if the MIR field gets too biased towards specific domains, systems and use cases?
3. Can a hype-biased scientific paper still be sufficiently fundamental and make a high and long-term impact?
4. Should a paper be fundamental in order to make impact?
5. What does "fundamental" mean?
6. How can we prevent that a hype-biased paper becomes obsolete after a few years?
7. Should hype-related publications be left to industrial R&D departments, since they are interested in realization and optimization of (their) specific products and services?
8. Should academia get more involved in addressing Internet hypes?
9. Why do we limit ourselves to only analyzing the hypes?
10. Could a new hype emerge from the fundamental insights obtained in the MIR community on how users access and manage multimedia content?

Moderator:
Alan Hanjalic, Delft University of Technology, The Netherlands

Panelists:
Nuria Oliver Ramirez, Telefonica R&D, Spain
Apostol Natsev, IBM Research
Alberto del Bimbo, University of Florence, Italy
Michael Lew, Leiden University, The Netherlands