[image 00520] 講演会のご案内（3月6日(木) 15:00-16:00@大阪府立大学 B4棟 西K-301室）

[Motoi Iwata]

Image-ML, Document-ML の皆様

大阪府立大学の岩田です。

Kaiserslautern大学の Tandra Ghose 教授をお迎えして、
下記の通り講演会を開催いたします。

ご興味がおありの方は是非ご参加ください。
事前登録等は一切不要です。
お知り合いにご興味をもたれる方がおられましたら、
お声をかけていただければ幸いです。

よろしくお願いいたします。

岩田　基
大阪府立大学 大学院 工学研究科
知能情報工学分野 第3グループ　助教
〒599-8531　大阪府堺市中区学園町1-1
TEL/FAX: 072-254-9281　　内線6805

-------------------------------------------------

日時　：2014年3月6日（木）　15:00-16:00
場所　：大阪府立大学　B4棟 3階 西K-301室
講師　：Prof. Dr. Tandra Ghose
        Department of Psychology, University of Kaiserslautern, Germany. 
http://www.sowi.uni-kl.de/fileadmin/wpsy/public/ghose.htm
主催：　大阪府立大学　文書解析・知識科学研究所 (IDAKS)

講演タイトル：
Studying Perceptual Tasks in Different Domains using Eye-Movements


講演概要：
Perception is the process of gleaning information about the world in 
order to interact with it in an optimal fashion. It includes deciding 
where to look, how we read or do photo editing, for example. Scientific 
investigation of perceptual processes include procedures, such as, 
conducting explicit interviews for recording phenomenological 
experiences or collecting manual response measures, such as, reaction 
time, percent correct responses, as a function of some change in 
stimulus property. These are coarse measures based on the final percept 
and do not provide a detailed sampling of the cognitive processes 
involved in acquiring information from the stimulus. Eye movement 
measures provide richer sampling of the process of information 
collection dependent on changing stimulus properties. I will present eye 
movement based studies in domains that are crucial for bridging the gap 
between human and computer vision. Firstly, I will present a study on 
development of an implicit measure based on saccadic metrics for 
strength of factors that bias perceptual grouping. Secondly, I will talk 
about an eye movement based study for investigating the 
information-content definition based on contextual probability that best 
predicts reading of sentences.  Finally, I will present some preliminary 
data for collecting implicit knowledge of photo editing experts. This 
technique supplements information from explicit interviews. Interviews 
do not provide sufficient information about features that can be 
directly used to train novices or develop software for image processing 
that models expert behavior. Such studies for extracting implicit 
features will help develop better technologies for human computer 
interaction.