Convis: A Toolbox To Fit and Simulate Filter-based Models of Early Visual Processing

Abstract

We developed Convis, a Python simulation toolbox for large scale neural populations which offers arbitrary receptive fields by 3D convolutions executed on a graphics card. The resulting software proves to be flexible and easily extensible in Python, while building on the PyTorch library (The Pytorch Project, 2017), which was previously used successfully in deep learning applications, for just-in-time optimization and compilation of the model onto CPU or GPU architectures. An alternative implementation based on Theano (Theano Development Team, 2016) is also available, although not fully supported. Through automatic differentiation, any parameter of a specified model can be optimized to approach a desired output which is a significant improvement over e.g., Monte Carlo or particle optimizations without gradients. We show that a number of models including even complex non-linearities such as contrast gain control and spiking mechanisms can be implemented easily. We show in this paper that we can in particular recreate the simulation results of a popular retina simulation software VirtualRetina (Wohrer and Kornprobst, 2009), with the added benefit of providing (1) arbitrary linear filters instead of the product of Gaussian and exponential filters and (2) optimization routines utilizing the gradients of the model. We demonstrate the utility of 3d convolution filters with a simple direction selective filter. Also we show that it is possible to optimize the input for a certain goal, rather than the parameters, which can aid the design of experiments as well as closed-loop online stimulus generation. Yet, Convis is more than a retina simulator. For instance it can also predict the response of V1 orientation selective cells. Convis is open source under the GPL-3.0 license and available from https://github.com/jahuth/convis/ with documentation at https://jahuth.github.io/convis/.

Retinal processing: insights from mathematical modelling

“…Eq. (33) expresses that drives are correlated due to the overlap of B cells RFs, a well known result. Especially, correlations decrease with the distance d between the two RFs center (like e −d 2 if RFs are Gaussian).…”

“…physically detailed models [26,27,28], Generalized Linear Models applied to retinal coding [29,30,31]. Several powerful software has been designed to model the retina at different scales such as COREM [32], Convis [33], Isetbio https: //github.com/isetbio/isetbio/wiki. The Virtual Retina simulator, developed by A. Wohrer and P. Kornprobst [34] at INRIA was one of the first of these simulators and has given rise to subsequent simulators in our group, the platform PRANAS [35], https://team.inria.fr/biovision/pranas-software/ and more recently Macular https://team.inria.fr/biovision/macular-software/.…”

“…What could be the origin of this discrepancy ? A first reason is that correlations in the retina are often thought in terms of the drive correlations (33). Reducing the overlap between B cells RFs, i.e.…”

“…Reducing the overlap between B cells RFs, i.e. decreasing the magnitude of the product (33) lowers the drive correlations. The idea is then that A cells lateral inhibition reduces the center part of the RF and increases the surround thereby reducing the RFs overlap.…”

“…In the rest of this section, we want to consider in more detail the effects of rectification on RG cells spike correlations. We want to show that they induce non stationary stimulus dependent correlations which are not due to the drives correlations (33).…”