Research

Modeling the development of topographic maps in the primary visual cortex

Primate vision research has shown that in the retinotopic map of the primary visual cortex, eccentricity and meridional angle are mapped onto two orthogonal axes: whereas the eccentricity is mapped onto the non temporal axis, the meridional angle is mapped onto the dorsoventral axis Theoretically such a map has been approximated by a complex log map. We attempt to model the development of the global retinotopic and orientation maps using a similar self-organizing model.

The architecture consists of a Laterally Interconnected Synergistically Self Organizing Map (LISSOM) with 2 layers; representing the retina, and the V1 respectively. At each time step, each neuron in V1, combines the afferent activation along with its lateral excitations and inhibitions from the previous time step. The afferent, lateral excitatory and lateral inhibitory weights adapt based on a normalized Hebbian mechanism.

Modeling the direction selectivity of V1 Cells in the primate visual cortex using neural field model

primary visual cortex, also known as V1 and Area 17, is part of the cerebral cortex and is the entry point of the visual information from the eye's retina. The cells in V1 are believed to respond to various features such as orientation, direction of motion, ocular dominance etc. Empirical studies provide quantitative evidence for the direction specificity of the responses of the cells in macaque monkey striate cortex. Many models of direction selectivity have been proposed and most of them assume linearity and process entire sequence at a single instance. On the other hand, experimental data show direction selectivity is non-linear in nature and non-linear excitation and inhibition from other neurons involved in the contribution to direction selectivity. Here explores biologically plausible dynamical system approach to model direction selectivity of V1 cells to variety of moving visual stimuli such as bars, edges and gratings, and correlated the results with the empirical reports. We showed that the Gabor like receptive fields were developed through learning from randomly initialized weights.

Modeling the pattern selectivity of cells in Middle Temporal Area

Visual area MT (Middle temporal), also known as area V5 is a region of extra striate cortex that is thought to play a major role in perception of motion, the integration of local motion signals into global percepts. MT receives its most important input from V1, and sends its major outputs to the areas located in the cortex immediately surrounding it, including areas FST, MST. Single unit analysis studies identified and categorized the population of cells in MT into two groups component selective and pattern selective. Pattern selective cells are capable of coding the motion of whole visual pattern independent of contours within them by overcoming the aperture problem. In this study we designed a simulation to test the hypothesis that the selectivity for the motion of whole patterns can be explained by the feedforward two-layer neural filed model proposed earlier. We tested our hypothesis using well known moving plaid sequences and naval moving 2D object (moving square sequence) and showed the simulation results are in consistent with the empirical reports.

Modeling optic flow selectivity of MSTd cells in the primate visual cortex

Extensive research on primate’s visual motion processing reports the cortical motion pathway is made of primary visual cortex (V1), middle temporal (MT) area and the Medial superior temporal (MST) area. The complex motion processing like optic flow is believed to be processed by the two areas MT and MST. MT contain many direction selective cells that in principle might form a distributed encoding of flow field arriving on the retina and the neurons in MSTd respond to large random dot optic flow patterns suggesting an involvement in the analysis of optic flow. The electrophysiological properties of neurons in MT showed that the large portion of cells were tuned to speed and direction of the moving stimuli. MSTd receives its primary input from MT where initial processing of optic flow taken place inform of calculation of direction and speed within the small region of visual field and these local MT motion estimates are used to achieve the global selectivity for optic flow by the neurons in MSTd. We present three models that are designed to simulate the computation of optic flow performed by the MST neurons in primates. The simulation results show that, while neurons in model-1 and model-2 could account for MSTd cell properties found neurobiologically, model-3 neuron responses are consistent with the idea of functional hierarchy in the macaque motion pathway. These results also suggest that the deep learning models could offer a computationally elegant and biologically plausible solution to simulate the development of cortical responses of the primate motion pathway.

Structure from motion and Biological motion processing

A breakthrough in the understanding of dynamic 3D shape recognition was the discovery that our visual system can extract 3D shape from inputs having only sparse motion cues such as (i) point light displays and (ii) random dot displays representing rotating 3D shapes - phenomena named as biological motion (BM) processing and structure from motion (SFM) respectively. We proposed dynamic deep network model to explain the mechanisms underlying both structure from motion (SFM) and biological motion (BM). By conducting different simulations to recognize 3D shape from rotating surfaces, we showed that smaller dot density of rotating shape, oriented shapes, occluding boundaries, and dynamic noise backgrounds reduced the model's performance whereas eliminating local feature stability, occluding intrinsic boundaries, and static noise backgrounds had little effect on shape recognition, suggesting that the motion of high curvature regions like shape boundaries provide strong cues in shape recognition. Through the simulations conducted on point light action sequences, we showed that critical joint movements and their movement pattern generated in the course of action (actor configuration) play a key role in action recognition performance.

Page updated

Google Sites

Report abuse