Invited speakers

Keynote speaker 1:

Nick Campbell
National Institute of Information and Communications Technology & ATR Spoken Language Communication Research Labs, Japan.

How to follow a conversation without listening to the words

Abstract:
This talk describes the use of nonverbal speech sounds in spoken interaction and shows how a technology can be trained to follow the flow of a conversation or dialogue without any understanding any of its propositional content. The first part of the talk describes the collection of a very large corpus of natural conversational speech, the second part outlines the major findings from its analysis, and the third part describes the development of a multimodal device for tracking participant status in such conversational dialogues.

Keynote speaker 2:

Vaclav Hlavac [www]
Center for Machine Perception, Czech Technical University, Prague

Structure and images

Abstract:
Statistical pattern recognition methods have had difficulties to deal with images for several decades. The main obstacle is that a standard statistical approach cannot directly cope with the structure induced by the neighborhood relation in images. The talk will demonstrate this issues on several examples. Many researchers think that attempts to apply structural pattern recognition methods in the 1960s and 1970s led to the dead end. I like to advocate that the structural pattern recognition can be embedded into the statistical pattern recognition framework. This step has the potential to bring robustness to the structural approach. There are several possibilities in which such approach solves practically applicable tasks. A few examples from our recent research will be given, e.g., (a) optimizations on Markovian random fields applied to non-rigid matching in images or segmentation; (b) the structural construction applied to grammar-based recognition of the on-line hand written text.