Faculty of Informatics
Computer-Human Interfaces
Personal information technologies
András Lőrincz
Neural Information Processing Group
http://nipg.inf.elte.hu
Eötv ö s L o rá n d U n iv e rsi ty F a c u lt y o f In fo rma tics Zsolt Ákos Marci Kati Melinda Dani Balázs II Balázs I Gergő Zoli I Gábor Gyula Szityú Viktor Zoli II András Company Research and academy Back from Columbia University Now, finishing
Thanks are due to my
to my group
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
1
• Main activites
2
• Personal information – example
3
• Typical & specific
4
• ―Dictionaries‖ and their integration
András Lőrincz http://nipg.inf.elte.hu
Content
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Main activites
Machine learning
Computer-human interaction (TÁMOP projects)
Natural language processing (US Air Force Research Lab)
Machine vision
Map-Reduce optimization (Morgan-Stanley)
Analysis of customers’ behavior (Hungarian Telekom)
Neuroscience & psychology learning from the structure of the brainEötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
1
• Main activites
2
• Personal information – example
3
• Typical & specific
4
• ―Dictionaries‖ and their integration
András Lőrincz http://nipg.inf.elte.hu
Content
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Motivation
Non-speaking, but speech understanding children with special needs
They can barely interact and communicate
Their intelligence, knowledge, and personality need interaction
augmentative and alternative communication
We develop enabling ICT tools for them
for communication and controlEötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
“Stone Age”
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Motivation
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics 9
Zozó and the first results
MonitorEötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics 10
The first three trials
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics 11
The sequence of 20 trials
cursor trajectories target positions on screen
Not perfect,
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics 12
Development: Bencus’ story
Highly imprecise cursor control: barely controlled head swings
In three months he learned how to control
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Motivation
Aibo webcam mike speakerEötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
It is for real
ultrasonic RF-MEMS distance estimation RF-MEMS motesLaptops and webcams
András Lőrincz http://nipg.inf.elte.hu
Practicing at the
Alternative and Augmentative Communication Center
Budapest
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Using Dasher
and
playing Load Balancing game
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Intelligent computer-human interface
Cognitive and emotional profiling
Intelligent dialogue system
Learning and development
virtual environments
edutainment
serious gamesWe / they need
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Data collection and example basedrecommendation systems
personalization
Interaction with other people
depersonalization
social networks, social games, social involvementWe / they also need
Viktor Gyenes
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Social Gaming
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
EU Future and Emerging Technology
Exhibition April 2009, Prague:
Science Beyond Fiction
webcam emotional monitoring gesturing
playing ―anger‖
―Not sufficient artificial intelligence‖
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Education
Two Classrooms
András Lőrincz http://nipg.inf.elte.hu Budapest Node
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Under planning
“Nurse HomeCare”
András Lőrincz http://nipg.inf.elte.hu Budapest Node
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Under planning
Elderly Apartment Houses:
“Gold Sunset”
Törökbálint
Újpalota
ZuglóAndrás Lőrincz http://nipg.inf.elte.hu Budapest Node
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
In sum
We need to collect large databases about
previous events (i.e., movies)
results of tests
development of interactions
emotions and cognitive profiles subject to
privacy
in order to have a predictive recommendation system
for typical events annotation through machine recognition
for specific events novelty recognition through machine learning
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
1
• Motivation
2
• Personal information – example
3
• Typical & specific
Recommendations
4
• Outlook:
• ―Dictionaries‖ and their integration
András Lőrincz http://nipg.inf.elte.hu
Content
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Dot patterns
July 08, 2010, Gatsby András Lőrincz http://nipg.inf.elte.hu
Basic forms
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Dot patterns
July 08, 2010, Gatsby András Lőrincz http://nipg.inf.elte.hu
Basic forms Study items
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Dot patterns
July 08, 2010, Gatsby András Lőrincz http://nipg.inf.elte.hu
Test items
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Dot patterns
July 08, 2010, Gatsby András Lőrincz http://nipg.inf.elte.hu
Test items
control
hippocampal subject
Test results on ―category learning‖
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Dot patterns
July 08, 2010, Gatsby András Lőrincz http://nipg.inf.elte.hu
Test results on recognition
control
hippocampal subject
Test results on ―category learning‖
control
hippocampal subject
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
normal child, age four years and two months
autistic child, age three and half years
Discrimination vs. generalization
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Autism
Individuals with autism have difficulty abstracting subtle spatial information that is necessary
for the formation of a mean prototype,
for categorizing faces and objects.Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Nine dots in autism
discrimination / visual acuity
Autistic group is less accurate
Number of prototypeformers is smaller
International Meeting for Autism Research May 7 - 9, 2009,
Gastgeb et al. Univ. Pittsburgh
Eagle-Eyed Visual Acuity Ashwin et al., Biol. Psych. 2008
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
1
• Main activites
2
• Personal information – example
3
• Typical & specific
4
• ―Dictionaries‖ and their integration
András Lőrincz http://nipg.inf.elte.hu
Content
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Route to dictionary integration
Annotated databases:
Wikipedia, WordNet, OpenCyc―ontologies‖
Map ontologies onto each other
―recommendation system‖
map typical parts
Map to other modalities
LabelMe: segmented images and textual information
segmented images and movies
model of the self: interaction—control—optimizationEötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics
Route to dictionary integration
Movies
Motion understanding and control
Images
Image understanding through texts
Texts, documents
Dialogue system
Eötv ö s L o rá n d U n iv e rsi F a c u lt y o f In fo rma tics