Date of Award
primary visual cortex, perceptual learning, contour integration, contour saliency
The earliest cortical stage of visual processing, the primary visual cortex, has long been seen as a static preprocessor that finds local edges and their orientation like a linear filter bank, and passes this information on to downstream visual areas. This view has been challenged in recent years since the discovery of contextual influences, that is, interactions between the responses of neurons that encode for non-overlapping adjacent areas of visual space, and their anatomical substrate, long-range horizontal connections. These contextual interactions have been shown in awake behaving primates to be modulated depending on the task the animals are performing. A first set of electrophysiological experiments has shown with the help of information theory that when an animal performed one of two tasks on the same visual display, the contextual modulations of the task-relevant parts of the visual display contained more information about the stimulus position than when the same elements were task-irrelevant. A second set of experiments on contour integration was analyzed with ROC analysis to show that an ideal observer could predict the presence of an embedded contour from the spike count of a single neuron on a single trial as well as the animalâ€™s behavioral performance. A final set of experiments showed that prior to learning the same contour integration task, the responses did not contain any information about the stimulus position, that the information in the response increased in parallel with the animals performance during learning, and that the enhanced response after learning disappeared during anesthesia, but is only weakened when performing an irrelevant task in a different part of visual space. Last, a neural network is presented that allows gating of long-range horizontal connections by top-down feedback. The stability and the dynamic behavior of the network have been established with phase-plane analysis. Large-scale simulations have been performed to confirm the stability and show the enhanced contour integration of realistic stimuli as a function of feedback gain. This model has fit quantitatively the electrophysiological experiments of contour integration.
Piëch, Valentin, "Perceptual Learning, Long-Range Horizontal Connections And Top-Down Influences In Primary Visual Cortex" (2009). Student Theses and Dissertations. 124.