Estimating psychophysical “perceptive fields” from human auditory experiments
The last decade has
seen considerable advances in the cognitive neurosciences on different
spatial scales, from molecular techniques to brain imaging. However,
computational models of perceptual and cognitive processes on a
behavioural level are still limited. Currently there are two obstacles
preventing the development of successful computational models of
cognitive processes: First and foremost is the feature identification
problem − which are the critical features, or cues, on which the human
sensory systems base their computations? For simple stimuli like points
of light or Gabor patches the experimenter typically forces the
observers to use the cues made available to them. For more complex
perceptual tasks, such as object recognition in natural scenes,
however, the cues are typically unknown. The second, and related,
obstacle is the limited applicability of the time-proven methods of
linear systems theory to perceptual and cognitive tasks. Both obstacles
may be overcome by a new “perceptive field” estimation technique we
developed over the last years, which makes use of modern machine
learning methods (kernel methods). Particularly helpful may prove our
non-linear systems identification approach thus far successfully
applied to predict visual saliency from eyemovement recordings.
In the proposed project these methods will be applied to auditory psychophysics, initially tonein-noise detection. Despite decades of research, no conclusive answer has been obtained to the fundamental question, which cues human observers employ to detect a pure tone presented in narrow-band background noise. Data suggest that observers may switch between multiple cues but research is hampered by a lack of knowledge about all the potential cues observers may use. Using our explorative non-linear system identification techniques should help us identify the critical auditory features, show how they may be selected depending on, e.g., task difficulty, guide new psychophysical experiments, and finally arrive at a computational model for this basic auditory ability believed to underlie many, if not most, auditory tasks.
From a purely computational point of view this is of interest as it extends our “perceptive field” estimation technique to the auditory domain where time is the critical dimension, perhaps helping development of spatio-temporal perceptive field estimation techniques for vision in the future.
In the proposed project these methods will be applied to auditory psychophysics, initially tonein-noise detection. Despite decades of research, no conclusive answer has been obtained to the fundamental question, which cues human observers employ to detect a pure tone presented in narrow-band background noise. Data suggest that observers may switch between multiple cues but research is hampered by a lack of knowledge about all the potential cues observers may use. Using our explorative non-linear system identification techniques should help us identify the critical auditory features, show how they may be selected depending on, e.g., task difficulty, guide new psychophysical experiments, and finally arrive at a computational model for this basic auditory ability believed to underlie many, if not most, auditory tasks.
From a purely computational point of view this is of interest as it extends our “perceptive field” estimation technique to the auditory domain where time is the critical dimension, perhaps helping development of spatio-temporal perceptive field estimation techniques for vision in the future.
| german version |