Life Sciences Commons^™

Virtual V1sion: A Collaborative Coding Project, Cheryl Olman

MODVIS Workshop

Virtual V1sion is a new idea for fostering modeling collaborations and data sharing. While still in its infancy, the ultimate goal is a website that hosts repositories for (1) interchangeable model elements, (2) datasets that can be fit/predicted by those models, and (3) educational modules that explain the background for both the models and the datasets. The scope of the modeling is limited to predictions of V1 responses, although not all computations represented by model elements in Virtual V1sion are required to be V1-intrinsic: a goal of the project is to provide a framework in which predictions for modulation by …

Parametrically Constrained Lightness Model Incorporating Edge Classification And Increment-Decrement Neural Response Asymmetries, Michael E. Rudd

MODVIS Workshop

Lightness matching data from disk-annulus experiments has the form of a parabolic (2^nd-order polynomial) function when matches are plotted against annulus luminance on log-log axes. Rudd (2010) has proposed a computational cortical model to account for this fact and has subsequently (Rudd, 2013, 2014, 2015) extended the model to explain data from other lightness paradigms, including staircase-Gelb and luminance gradient illusions (Galmonte, Soranzo, Rudd, & Agostini, 2015). Here, I re-analyze parametric lightness matching data from disk-annulus experiments by Rudd and Zemach (2007) and Rudd (2010) for the purpose of further testing the model and to try to constrain …

Failure Of Surface Color Cues Under Natural Changes In Lighting, David H. Foster, Iván Marín-Franch

MODVIS Workshop

Color allows us to effortlessly discriminate and identify surfaces and objects by their reflected light. Although the reflected spectrum changes with the illumination spectrum, cone photoreceptor signals can be transformed to give useful cues for surface color. But what happens when both the spectrum and the geometry of the illumination change, as with lighting from the sun and sky? Is it possible, as a matter of principle, to obtain reliable cues by processing cone signals alone? This question was addressed here by estimating the information provided by cone signals from time-lapse hyperspectral radiance images of five outdoor scenes under natural …

Modeling The Joint Distribution Of Scene Events At An Edge, James Elder, Ying Li

MODVIS Workshop

Edges in an image arise from discontinuities in scene variables, namely reflectance (R), illumination (I), depth (D) and surface orientation (O). Prior studies on edge classification have viewed it as a binary classification problem: each edge is assumed to arise from one of two disjoint categories (e.g., depth or not depth, shadow or not shadow). Here we suggest an alternate view in which an edge may signal discontinuities in any combination of the scene variables (RIDO). To explore this model, we had 4 trained observers label one randomly selected edge in each of 1,000 randomly selected images drawn from the …

Identifying Falsifiable Predictions Of The Divisive Normalization Model Of V1 Neurons, Tadamasa Sawada, Alexander A. Petrov

MODVIS Workshop

The divisive normalization model (DNM, Heeger, 1992) accounts successfully for a wide range of phenomena observed in single-cell physiological recordings from neurons in primary visual cortex (V1). The DNM has adjustable parameters to accommodate the diversity of V1 neurons, and is quite flexible. At the same time, in order to be falsifiable, the model must be rigid enough to rule out some possible data patterns. In this study, we discuss whether the DNM predicts any physiological result of the V1 neurons based on mathematical analysis and computational simulations. We identified some falsifiable predictions of the DNM. The main idea is …

Modelling Response Properties Across The Orientation Map In Visual Cortex, Erin M. Koch, Jianzhong Jin, Jose-Manuel Alonso, Qasim Zaidi

MODVIS Workshop

Stimulus orientation in the primary visual cortex of primates and carnivores is mapped as iso-orientation domains radiating from pinwheel centers, where orientation preferences of neighboring cells change circularly. Whether this orientation map has a function is debated, because many mammals, such as rodents, do not have such maps. Here we model our physiological results that two fundamental properties of visual cortical responses, contrast saturation and cross-orientation suppression, are stronger within iso-orientation domains than at pinwheel centers. Our model expands on a standard thalamic model of cross orientation suppression, and explains differences between orientation domains by intra-cortical excitation (not normalization) from …

Derivatives And Inverse Of A Linear-Nonlinear Multi-Layer Spatial Vision Model, Borja Galan, Marina Martinez-Garcia, Praveen Cyriac, Thomas Batard, Marcelo Bertalmio, Jesus Malo

MODVIS Workshop

Analyzing the mathematical properties of perceptually meaningful linear-nonlinear transforms is interesting because this computation is at the core of many vision models. Here we make such analysis in detail using a specific model [Malo & Simoncelli, SPIE Human Vision Electr. Imag. 2015] which is illustrative because it consists of a cascade of standard linear-nonlinear modules. The interest of the analytic results and the numerical methods involved transcend the particular model because of the ubiquity of the linear-nonlinear structure.

Here we extend [Malo&Simoncelli 15] by considering 4 layers: (1) linear spectral integration and nonlinear brightness response, (2) definition of local contrast …

Towards A Functional Explanation Of The Connectivity Lgn - V1, Marina Martinez-Garcia, Borja Galan, Luis M. Martinez, Jesus Malo

MODVIS Workshop

The principles behind the connectivity between LGN and V1 are not well understood. Models have to explain two basic experimental trends: (i) the combination of thalamic responses is local and it gives rise to a variety of oriented Gabor-like receptive felds in V1 [1], and (ii) these filters are spatially organized in orientation maps [2]. Competing explanations of orientation maps use purely geometrical arguments such as optimal wiring or packing from LGN [3-5], but they make no explicit reference to visual function. On the other hand, explanations based on func- tional arguments such as maximum information transference (infomax) [6,7] usually …

Towards A Unified Model Of Classical And Extra-Classical Receptive Fields, David A. Mély, Thomas Serre

MODVIS Workshop

One of the major goals in neuroscience is to understand how the cortex processes information. A substantial effort has thus gone into mapping classical receptive fields (cRF) across areas of the visual cortex and characterizing input-output relationships through linear-nonlinear response functions. Recently, there has been a lot of interest in mapping the extra-classical receptive field (extra-cRF) as well, by using contextual stimuli. The extra-cRF is a region outside the cRF that modulates a cell’s response but that is incapable of driving it on its own. However, existing models typically focus on one particular visual modality (form, motion, disparity or color), …

How Deep Is The Feature Analysis Underlying Rapid Visual Categorization?, Sven Eberhardt, Jonah Cader, Thomas Serre

MODVIS Workshop

Rapid categorization paradigms have a long history in experimental psychology: Characterized by short presentation times and fast behavioral responses, these tasks highlight both the speed and ease with which our visual system processes natural object categories. Previous studies have shown that feed-forward hierarchical models of the visual cortex provide a good fit to human visual decisions. At the same time, recent work has demonstrated significant gains in object recognition accuracy with increasingly deep hierarchical architectures: From AlexNet to VGG to Microsoft CNTK – the field of computer vision has championed both depth and accuracy. But it is unclear how well …

Spatial Synaptic Growth And Removal For Learning Individual Receptive Field Structures, Michael Teichmann, Fred H. Hamker

MODVIS Workshop

One challenge in creating neural models of the visual system is the appropriate definition of the connectivity. The modeler constrains the results with its definition. Unfortunately, there is often just insufficient information about connection sizes available, e.g. for deeper layer or different neuron types like interneurons. Hence, a mechanism refining the connection structure based on the learnings would be appreciated.

Such mechanism can be found in the human brain by structural plasticity. That is, the formation and removal of synapses. For our model, we exploit that synaptic connections are likely to be formed in the proximity of other synapses and …

A Geometric Approach To Sparse Coding Yields Insight Into Nonlinear Responses, Kedarnath Vilankar, James Golden, David Field

MODVIS Workshop

In artificial and biological networks, it is a common accepted practice to describe a neurons (biological or artificial) response properties by a two-dimensional feature map (receptive field). However, real neurons have nonlinear response properties which are not represented by their receptive fields. The efficient coding mechanisms such as sparse coding network or ICA, learn the response properties of V1 neurons from natural images using neural networks. These networks learn the receptive fields which are similar to the receptive fields of V1 neurons. These networks also produces some of the nonlinearities (such as end-stopping and non-classical surround effect), which are exhibited …

Using Deep Features To Predict Where People Look, Matthias Kümmerer, Matthias Bethge

MODVIS Workshop

When free-viewing scenes, the first few fixations of human observers are driven in part by bottom-up attention. We seek to characterize this process by extracting all information from images that can be used to predict fixation densities (Kuemmerer et al, PNAS, 2015). If we ignore time and observer identity, the average amount of information is slightly larger than 2 bits per image for the MIT 1003 dataset. The minimum amount of information is 0.3 bits and the maximum 5.2 bits. Before the rise of deep neural networks the best models were able to capture 1/3 of this information on average. …

Modelling Short-Latency Disparity-Vergence Eye Movements Under Dichoptic Unbalanced Stimulation, Agostino Gibaldi, Guido Maiello, Peter J. Bex, Silvio P. Sabatini

MODVIS Workshop

Vergence eye movements align the optical axes of our two eyes onto an object of interest, thus facilitating the binocular summation of the images projected onto the left and the right retinae into a single percept. Both the computational substrate and the functional behaviour of binocular vergence eye movements have been the topic of in depth investigation. Here, we attempt to bring together what is known about computation and function of vergence mechanism. To this aim, we evaluated of a biologically inspired model of horizontal and vertical vergence control, based on a network of V1 simple and complex cells. The …

Learning Object Representations For Modeling Attention In Real World Scenes, Alex Schwarz, Frederik Beuth, Fred H. Hamker

MODVIS Workshop

Models of visual attention have been rarely used in real world tasks as they have been typically developed for psychophysical setups using simple stimuli. Thus, the question remains how objects must be represented to allow such models an operation in real world scenarios. We have previously presented an attention model capable of operating on real-world scenes (Beuth, F., and Hamker, F. H. 2015, NCNC, which is a successor of Hamker, F. H., 2005, Cerebral Cortex), and show here how its object representations have been learned. We have used a learning rule based on temporal continuity (Földiák, P., 1991, Neural Computation) …