Keywords
visual attention, object localisation, visual search, temporal continuity learning, figure-ground separation
Abstract
Models of visual attention have been rarely used in real world tasks as they have been typically developed for psychophysical setups using simple stimuli. Thus, the question remains how objects must be represented to allow such models an operation in real world scenarios. We have previously presented an attention model capable of operating on real-world scenes (Beuth, F., and Hamker, F. H. 2015, NCNC, which is a successor of Hamker, F. H., 2005, Cerebral Cortex), and show here how its object representations have been learned. We have used a learning rule based on temporal continuity (Földiák, P., 1991, Neural Computation) to ensure biological plausibility. Yet, temporal continuity learning rules have not been used in a real world context, thus, we conducted an improvement: We increased the postsynaptic threshold to make the learning more specific, resulting in object-encoding cells reacting mainly specific for their preferred objects. Furthermore, we present a novelty in relation to Beuth, F. and Hamker, F. H., 2015: the learning of object representation invariant towards the background. It is currently unknown how such representations are learned by the human brain. Suggestions have been made to use disparity or motion, whereas we propose temporal continuity learning. This principle learns connections from presynaptic features which are stable over time. As the object changes much less than the background over time, strong connections are primarily learned to the object and no connections to the background. Such learned representations allow the attention model to identify and locate objects in real world scenes.
Start Date
11-5-2016 10:00 AM
End Date
11-5-2016 10:25 AM
Location
Chemnitz, Germany
Learning Object Representations for Modeling Attention in Real World Scenes
Chemnitz, Germany
Models of visual attention have been rarely used in real world tasks as they have been typically developed for psychophysical setups using simple stimuli. Thus, the question remains how objects must be represented to allow such models an operation in real world scenarios. We have previously presented an attention model capable of operating on real-world scenes (Beuth, F., and Hamker, F. H. 2015, NCNC, which is a successor of Hamker, F. H., 2005, Cerebral Cortex), and show here how its object representations have been learned. We have used a learning rule based on temporal continuity (Földiák, P., 1991, Neural Computation) to ensure biological plausibility. Yet, temporal continuity learning rules have not been used in a real world context, thus, we conducted an improvement: We increased the postsynaptic threshold to make the learning more specific, resulting in object-encoding cells reacting mainly specific for their preferred objects. Furthermore, we present a novelty in relation to Beuth, F. and Hamker, F. H., 2015: the learning of object representation invariant towards the background. It is currently unknown how such representations are learned by the human brain. Suggestions have been made to use disparity or motion, whereas we propose temporal continuity learning. This principle learns connections from presynaptic features which are stable over time. As the object changes much less than the background over time, strong connections are primarily learned to the object and no connections to the background. Such learned representations allow the attention model to identify and locate objects in real world scenes.