These days, an enormous amount of multimedia data is available on various kinds of Web sites and devices. Until now, R&D of statistical speech processing has been focused on high-quality data annotation and parsimonious model construction using the annotated data. However, from now on, the R&D focus will shift to the issue of how to construct a model that is robust against diverse types of noise in a massive amount of data annotated with either no labels or only unreliable ones. Another subject that will receive attention is how to convert domain knowledge based on a massive amount of data into model construction in different domains that have sparse data, e.g., for speech recognition systems for rare languages with few data resources. In such R&D, it is difficult to use a large amount of data from the beginning and it is necessary to investigate scalable methods that suit various amounts and quality-levels of data and domain knowledge. In this workshop, considering the present circumstances, researchers in machine learning and in speech, natural language, and image processing will get together and discuss scalable approaches in the era of abundant data.

Location
Kyoto International Conference Center, Kyoto, Japan
http://www.icckyoto.or.jp/en/index.html

Important dates
20/Jan/2012 Extended abstract submission deadline
20/Feb/2012 Notification of acceptance
31/Mar/2012 Workshop date

Language
The working language of the conference is English.

Support
Network of Excellence for Statistical Machine Learning, Institute of Statistical Mathematics