Selim Onat

My main occupation is Neuroscience.

In the past, I have been interested on how the visual system processes natural scenes. To this end, I have recorded naturalistic movies using micro-cameras carried by cats while they were actively exploring a natural environment. I used these movies to train neuronal networks in an unsupervised manner and compared learnt features to the known properties of neurons in visual cortex. I also used these videos as stimuli during physiological recordings to gain insights on the principles of natural signal processing in the visual cortex.

Recently, I started working on how humans make generalizations based on what they have previously learnt. To this end, I am using a variety of methodologies including fMRI (1), autonomous (2), as well as eye-movement recordings (3). This research emanates from the well-established field of "stimulus generalization" following mainly the "lineage" of Hovland, Hull and Roger Shepard (4), and including the more recent computational work of Josua Tenenbaum (5). Furthermore, it integrates work on anxiety disorders, as it is believed that these mechanisms are impaired in people suffering from anxiety problems.

Representation of Simple Stimuli across the Visual Cortex

Moving edges, which consist of drifting dark and bright bars, are kind of stimuli that are commonly used in physiological experiments mainly because of their parametrically controllable aspects.

The video below shows the large-scale cortical dynamics at the spatial scale of several millimeters during presentation of such visual stimuli.

Overlaid on the well-known orientation maps (shown at the bottom row), which are evoked by the specific orientation of the stimuli, drifting edges are furthermore represented by propagating waves (top row). The original publication can be found here.


NeuroImage from sonat on Vimeo.

Caption: Multiplexing of space and orientation information. The data presented in Fig. 1 and 2 is presented as a video. Upper video: Propagating activity reconstructed by combining oscillatory SVD components, averaged across several propagation cycles during stimulus presentation (cf. Fig. 3 and Fig. 4a). M = Medial, P = Posterior. Lower video: Propagating waves are shown in combination with tonic SVD components representing the orientation maps. The weight of both components were equalized prior to their combination. Contour lines are drawn at 90th activity percentiles of the tonic components.


Decomposition of evoked cortical responses to gratings of 0.2 c/deg drifting for 2 s at a temporal frequency of 6.25 Hz. (a) Evoked spatio-temporal activity patterns (top rows) and time courses obtained by spatial averages across the images (bottom traces) expressed as fractional change in fluorescence relative to blank condition (ΔF/F). Top left frame shows the vascular image of the recorded right hemisphere, P = posterior, L = lateral; here and in all figures scale bar 1 mm. Leftmost frame in 2nd row depicts the time-averaged orientation map derived by subtracting evoked responses to the vertical grating from horizontal. Green trace = responses to vertical grating, drifting rightwards in visual space; blue trace = horizontal grating, drifting downwards. (b) Top left corner, singular values, gi, ranked in order of their contributions. Components of significant contribution to variance are colored (gray area depicts significance level). The contribution of each single SVD component to single recorded trials (n=35) was computed, their correlations across trials are represented as a matrix. Spatial (ui(x)) and temporal (vi(t)) modes of the SVD components were clustered according to their correlation (red, yellow, and green boxes; curves represent weight of each spatial mode [y-axes] as a function of time [400–1800 ms]).