|
Neural Computation,
Computer Vision, Robotics
Interactions of Intrinsic and Synaptic Plasticity
Neural Computation/ Development
Biological neurons are highly adaptive computation devices. In
addition to synaptic plasticity mechanisms, neurons also change
their non-linear behavior by adapting their membrane properties.
Such intrinsic plasticity is still poorly understood. In this
project, we develop models of intrinsic plasticity mechanisms and how they may
synergistically interact with synaptic plasticity mechanisms for
the learning of efficient neural codes.
To read more... Publications: Triesch (2004)
Back to Top
Autonomous learning of visual skills during development
Neural Computation/ Development/ Robotics/ Computer
Vision
In this project we build robotic and virtual models of how infants
might progressively acquire knowledge about the visual
world and in particular human faces and social
interactions. Topics covered range from autonomous learning of
accurate gaze shifts, over autonomous learning of object
representations, to studies of the emergence of joint attention.
The latter effort is part of the MESA
project and more details can be found there.
To read more... Publications: Triesch (2001)
Back to Top
"Democratic" Integration of visual cues
Neural Computation/Computer Vision
In this project we propose a method for integrating different visual
cues (like motion, color, shape, etc.) in a self-organized manner. The
method leads to automatic suppression and re-calibration of discordant
cues. The overall system shows an extremely high robustness with
respect to unforseen changes in the environment in contrast to
classical statistical approaches. We are currently investigating
different ways of extending this approach into a hierarchical
architecture. We also try to understand how multiple adaptive
visual cues might spontaneously collaborate in a self-organized fashion
to achieve novel perceptual goals that the system is facing for the
first time.
To read more...Publications:
Triesch&v.d.Malsburg (2001), Triesch (2000),
Triesch&v.d.Malsburg(2000)
Back to Top
Biologically Inspired Object Recognition Architectures
Computer Vision/Neural Computation/Robotics
In this project we develop neurally inspired
object recognition architectures for the analysis of complex natural
scences with clutter and occlusions. Topics include the sharing of
multiple features for efficient, scalable recognition of large numbers of objects, the integration of various feature types in a probabilistic framework,
the benefits of hierarchical recognition architectures inspired by the
primate visual cortex, and the coupling of recognition and
segementation into a closed feedback loop in such architectures.
To read more...Publications:
Murphy-Chutorian and Triesch (2005), Triesch and Eckes (2004), Triesch and v.d.Malsburg (2001b).
Back to Top
Virtual Reality and
Visual Psychophysics
Most of our psychophysics projects are in
collaboration with Mary Hayhoe
and Dana
Ballard at the
Virtual Reality Laboratory at the University of Rochester.
Task-Dependent Dynamics of Visual Working Memory
Visual Psychophysics
This project investigates how visual working memory represents natural
scenes. We hypothesize that a subject's moment-to-moment visual
representations are governed by their current task. We are testing this
hypothesis using a dual-task paradigm, which requires subjects to
detect changes while also engaging in a separate information gathering
task. Using this technique, we hope to determine how different
information gathering tasks modulates the types of changes subjects
notice most readily.
To read more...Project Website
Back to Top
Change Blindness in Natural Tasks
Visual Psychophysics/ Virtual Reality
In this project that is performed in collaboration with the
Virtual Reality Lab of the University of Rochester we study the
phenomenon of change blindness in natural tasks using state of the art
virtual reality equipment including haptic force feedback devices. We
could show that the ability of subjects to notice changes to the
central object of interest that occur across saccades is strongly
influenced by the task relevance of the changed object feature. We
could also demonstrate that this effect is not due to a different use
of gaze but is rooted in differences in internal
processing.
To read more...Publications: Triesch et al.
(2003), Hayhoe et.al. (2002), Triesch et.al. (2001)
Back to Top
Eye tracking technology in Virtual Reality
Virtual Reality/Visual Psychophysics
In this project that is also performed in collaboration with the
Virtual Reality Lab of the University of Rochester we investigate
methods for saccade contingent updating in virtual reality, i.e. the
technique of making a change to the displayed scene while a saccade of
the observer is progressing. This requires robust detection of saccade
onsets with very low latencies. Conventional eye tracking techniques
are either too slow or cannot be applied in Virtual Reality,
where subjects wear a head mounted display. We good very good results
by integrating two complementary eye tracking devices into a virtual
reality helmet.
To read more...Publications:
Triesch, Sullivan, Hayhoe (2002).
Back to Top
Fast Dynamics of Visual Cue Integration
Visual Psychophysics/Virtual Reality
In this project we studied in how far the human visual system is
capable of rapidly re-weighting different cues as suggested by the
Democratic Integration hypothesis mentioned above. We found that
subjects engaging in an object tracking task can discount a cue that
becomes unreliable in less than a second.
To read more...Publications:
Triesch, Ballard, Jacobs (2001, 2002).
Back to Top
|
