Poster Presentation "Computational Modeling of Optic Flow Selectivity in MSTd Neurons"

Purpose. To examine the development of selectivity and receptive field size in MSTd neurons sensitive to optic flows using neural network models with biologically realistic units. Methods. A two-layer back propagation network was used with inputs consisted of 1072 MT responses to radial, circular, and translational stimuli with varying signal-to-noise ratios. Sixty seven overlapping MT receptive fields were placed pseudo-randomly in the MSTd receptive field such that each corresponded to 16 directionally selective MT neurons which equally divided the vector space. Hidden units were classified as MSTd neurons whose receptive fields encompassed the MT receptive fields from the input layer. The output layer consisted of MSTd neurons whose receptive fields were coincident with those of the hidden layer. The neurophysiological selectivities in MSTd to optic flow stimuli were simulated in the output layer to examine their effects on the development of optic flow selectivity in the hidden layer. The effects of receptive field sizes in MSTd on network responses were examined using psychophysical stimuli outlined by Morrone, Burr, and Vaina (1995). Results. 1) The hidden units developed gaussian response profiles to optic flow stimuli with uniformly distributed means and sav= 60 ± 38 deg. 2) The hidden unit responses and degree of position invariance were proprtional to stimulus size. 3) Hidden unit position invariance decreased when tested with suboptimal stimuli. 4) In all simulations, the hidden units developed a continuum of optic flow selectivities regardless of the biases associated with the specification of output unit selectivities to optic flow. We will also report results obtained using more biologically realistic architectures. Conclusion. The hidden units developed selectivities to optic flow stimuli consistent with neurophysiological and psychophysical results. The inability to bias hidden unit development suggests a continuum of optic flow selectivities may be an efficient encoding of visual motion components of optic flow stimuli.
 

Morrone M. C., Burr D. C. & Vaina L. M. (1995). "Two stages of visual processing for radial and circular motion", Nature 376, 507-509.