Neural models of learning and visual grouping in the presence of finite conduction velocities
The hypothesis of object binding-by-synchronization in the visual cortex has been supported by recent experiments in awake monkeys. They demonstrated coherence among gamma-activities (30–90 Hz) of local neural groups and its perceptual modulation according to the rules of figure-ground segregation....
Keine Tags, Fügen Sie den ersten Tag hinzu!
|Zusammenfassung:||The hypothesis of object binding-by-synchronization in the visual cortex has been supported by recent experiments in awake monkeys. They demonstrated coherence among gamma-activities (30–90 Hz) of local neural groups and its perceptual modulation according to the rules of figure-ground segregation. Interactions within and between these neural groups are based on axonal spike conduction with finite velocities. Physiological studies confirmed that the majority of transmission delays is comparable to the temporal scale defined by gamma-activity (11–33 ms). How do these finite velocities influence the development of synaptic connections within and between visual areas? What is the relationship between the range of gamma-coherence and the velocity of signal transmission? Are these large temporal delays compatible with recently discovered phenomenon of gamma-waves traveling across larger parts of the primary visual cortex? The refinement of connections in the immature visual cortex depends on temporal Hebbian learning to adjust synaptic efficacies between spiking neurons. The impact of constant, finite, axonal spike conduction velocities on this process was investigated using a set of topographic network models. Random spike trains with a confined temporal correlation width mimicked cortical activity before visual experience. After learning, the lateral connectivity within one network layer became spatially restricted, the width of the connection profile being directly proportional to the lateral conduction velocity. Furthermore, restricted feedforward divergence developed between neurons of two successive layers. The size of this connection profile matched the lateral connection profile of the lower layer neuron. The mechanism in this network model is suitable to explain the emergence of larger receptive fields at higher visual areas while preserving a retinotopic mapping. The influence of finite conduction velocities on the local generation of gamma-activities and their spatial synchronization was investigated in a model of a mature visual area. Sustained input and local inhibitory feedback was sufficient for the emergence of coherent gamma-activity that extended across few millimeters. Conduction velocities had a direct impact on the frequency of gamma-oscillations, but did neither affect gamma-power nor the spatial extent of gamma-coherence. Adding long-range horizontal connections between excitatory neurons, as found in layer 2/3 of the primary visual cortex, increased the spatial range of gamma-coherence. The range was maximal for zero transmission delays, and for all distances attenuated with finite, decreasing lateral conduction velocities. Below a velocity of 0.5 m/s, gamma-power and gamma-coherence were even smaller than without these connections at all, i.e., slow horizontal connections actively desynchronized neural populations. In conclusion, the enhancement of gamma-coherence by horizontal excitatory connections critically depends on fast conduction velocities. Coherent gamma-activity in the primary visual cortex and the accompanying models was found to only cover small regions of the visual field. This challenges the role of gamma-synchronization to solve the binding problem for larger object representations. Further analysis of the previous model revealed that the patches of coherent gamma-activity (1.8 mm half-height decline) were part of more globally occurring gamma-waves, which coupled over much larger distances (6.3 mm half-height decline). The model gamma-waves observed here are very similar to those found in the primary visual cortex of awake monkeys, indicating that local recurrent inhibition and restricted horizontal connections with finite axonal velocities are sufficient requirements for their emergence. In conclusion, since the model is in accordance with the connectivity and gamma-processes in the primary visual cortex, the results support the hypothesis that gamma-waves provide a generalized concept for object binding in the visual cortex.|