Odelia Schwartz, PhD
Principal Investigator

odelia.schwartz [at] einstein.yu.edu

I employ tools of computational and theoretical neuroscience, to study sensory systems from the neural level through to perception and behavior. A main focus has been building models of neural processing based on the hypothesis that images, movies, and sounds have predictable and quantifiable statistical regularities to which the brain is sensitive. For a description of some current projects in the lab, click here.

Ruben Coen Cagli, PhD
PostDoc

+ 1.718.430.3040
ruben.coencagli [at] einstein.yu.edu

I study spatial context in images and in early vision. I am currently working on: a) developing a hierarchical generative model of statistical dependencies in natural images, and apply it to model the nonlinear surround modulation in the primary visual cortex; b) in the framework of such model, unify neural and perceptual phenomena due to spatial context in natural images (in collaboration with Peter Dayan, Gatsby UCL) [1,2]; c) testing detailed model predictions on early-cortical neural responses to natural images (in collaboration with Adam Kohn, AECOM); d) extending the hierarchical model to higher-level image features, and extrastriate cortical neurons.
I received the PhD degree in Physics in december 2007 from the University of Napoli, Italy. My doctoral research exploited eye tracking experiments and Bayesian modeling to address visuomotor coordination in the activity of drawing [3], and to develop an artificial agent with such capabilities [4].

[1] R.C.C., P. Dayan, O. Schwartz (2012) Cortical Surround Interactions and Perceptual Salience Via Natural Scene Statistics. PLoS Computational Biology, in press.
[2] R.C.C., P. Dayan, O. Schwartz (2009) Statistical Models of Linear and Nonlinear Contextual Interactions in Early Visual Processing. Advances in Neural Information Processing Systems, 22.
[3] R.C.C., P. Coraggio, P. Napoletano, O. Schwartz, M. Ferraro, G. Boccignone (2009) Visuomotor Characterization of Eye Movements in a Drawing Task. Vision Research, 49, 810-818.
[4] R.C.C., P. Coraggio, P. Napoletano, G. Boccignone, A.De Santis (2008) Sensorimotor coupling via Dynamic Bayesian Networks. Proceedings of the IEEE International Conference on Robotics and Automation 2008, ICRA.

Florian Roehrbein, PhD
PostDoc

+ 1.718.430.3040
florian.roehrbein [at] einstein.yu.edu

Statistical properties of natural scenes are usually analyzed based on random samples from static images. In order to arrive at a more biological setting, the selective human gazing behavior has to be taken into account. In this project I focus on video clips since they seem to be much more natural, though much rarely studied. See here some examples of the sort of stimuli I use in my experiments [firefox + safari].

We take the notion of natural images and natural sequences quite serious and thus do not allow for any human-made objects in our stimulus material. They also contain no camera movements or any kind of cuts, zooming etc.

Please contact me if you are interested in the natural video database I've collected. The videos are partly based on selected clips of the DynTex database (but transformed to greyscale and uniform duration).

Our group has previously applied generative models of scene statistics (e.g., Gaussian Scale Mixture model) to learn the joint statistical dependencies of oriented filter activations in random patches of static scenes without eye movements. We are now in the process of applying this model to determine the patterns of dependencies in our behaviorally relevant gaze-selected patches of image sequences.

For more information on all my research interests please click here.

Michoel Snow
MD PhD Student

+ 1.718.430.3040
michoel.snow [at] einstein.yu.edu

Over time, repeated exposure to relatively constant visual stimuli results in decreased neuronal response, i.e. adaptation. I am working to model this phenomenon using the Gaussian Scale Mixture (GSM) model. This model allows us to decompose any stimuli into a gaussian and a mixer variable. In natural scenes, while the gaussian changes from frame to frame, the mixer variable may stay constant, change slightly or very rapidly. By analyzing which of the states (constant, slow shift, or rapid shift) each scene falls into we hope to predict the neuronal response to that scene.

Jean Masterson
2009 rotation student

Nick Fernandez
2008 rotation student

Kyoko Kamishima
2008 rotation student