Purchase this article with an account.
Stefan Ringbauer, Florian Raudies, Heiko Neumann; Local Perceptual Learning for Motion Pattern Discrimination: a Neural Model. Journal of Vision 2010;10(7):1110. doi: https://doi.org/10.1167/10.7.1110.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Problem. Perceptual learning increases the performance of motion pattern discrimination (Nishina et al., J.of Vision 2009). The results suggest that local, not global learning mechanisms gained the improvement. The question remains which mechanisms of cortical motion processing are involved and how neural mechanisms of learning can account for this achievement. Method. We build upon a neural model for motion and motion pattern detection that incorporates major stages of the dorsal pathway, namely areas V1, MT, and MSTd which has been extended by a stage for decision-making in area LIP. MSTd cells are sensitive to patterns of motion by integrating motion direction sensitive MT activities along the convergent feedforward signal pathway. Feedback from MSTd to MT neurons modulates their activity. The strength of connection weights between MT and MSTd neurons can be adapted by repetitive presentation of motion patterns. MSTd to MT feedback also modulates the weight adaptation process by employing a variant of Hebbian learning using Oja's rule (J.Math.Biology 1982). As a consequence MT cell tuning changes and in turn improves the discrimination performance of perceived motion patterns. Results and Conclusion. Model simulations quantitatively replicate the findings of Nishina and co-workers. Specifically, discrimination learning between target and neutral pattern improves from d′=0.075 to d′=0.113. The model predicts that the presentation of rotation patterns leads to the same performance as for the radial motion patterns. In addition, our computational simulations suggest that decision performance as well as the threshold differences for motion discrimination drop if noise is added to the visual stimulus. Our model predicts that feedback from area MSTd to MT stabilizes the learning under conditions when noise significantly impairs the coherence of the input motion. This suggests that while the perceptual learning in this case might indeed be local, more global information is involved for stabilizing the learning process.
This PDF is available to Subscribers Only