A model for the development of simple cell receptive fields and the ordered arrangement of orientation columns through activity-dependent competition between ON- and OFF-center inputs

By Kenneth D. Miller. 1994

Neurons in the primary visual cortex of higher mammals respond selectively to light/dark borders of a particular orientation. The receptive fields of simple cells, a type of orientation-selective cell, consist of adjacent, oriented regions alternately receiving ON-center and OFF-center excitatory input. I show that this segregation of inputs within receptive fields can occur through an activity-dependent competition between ON-center and OFF-center inputs, just as segregation of inputs between different postsynaptic cells into ocular dominance columns appears to occur through activity-dependent competition between left-eye and right-eye inputs. These different outcomes are proposed to result, not from different mechanisms, but from different spatial structures of the correlations in neural activity among the competing inputs in each case. Simple cells result if ON-center inputs are best correlated with other ON-center inputs, and OFF with OFF, at small retinotopic separations, but ON-center inputs are best correlated with OFF-center inputs at larger separations. This hypothesis leads robustly to development of simple cell receptive fields selective for orientation and spatial frequency, and to the continuous and periodic arrangement of preferred orientation across the cortex. Input correlations determine the mean preferred spatial frequency and degree of orientation selectivity. Estimates of these correlations based on measurements in adult cat retina (Mastronarde, 1983a,b) produce quantitative predictions for the mean preferred spatial frequencies of cat simple cells across eccentricities that agree with experiments (Movshon et al., 1978b). Intracortical interactions are the primary determinant of cortical organization. Simple cell spatial phases can play a key role in this organization, so arrangements of spatial phases and preferred orientations may need to be studied together to understand either alone. Possible origins for other cortical features including spatial frequency clusters, afferent ON/OFF segregation, blobs, pinwheels, and opponent inhibition within simple cell receptive fields are suggested. A number of strong experimental tests of the hypothesis are proposed.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Deep learning in neural networks: An overview (2014, background): “…1990; Földiák & Young, 1995; Grossberg, 1976a, 1976b; Hebb, 1949; Kohonen, 1972, 1982, 1988; Kosko, 1990; Martinetz, Ritter, & Schulten, 1990; Miller, 1994; Mozer, 1991; Oja, 1989; Palm, 1992; Pearlmutter &Hinton, 1986; Ritter & Kohonen, 1989; Rubner & Schulten, 1990; Sanger, 1989; Saund,…”
Predictive Feedback Can Account for Biphasic Responses in the Lateral Geniculate Nucleus (2009, background): “In addition, a variety of models has been proposed to account for orientation selectivity in early visual cortex [9,43–45].”
Mechanisms for Stable, Robust, and Adaptive Development of Orientation Maps in the Primary Visual Cortex (2013, methodology): “Unlike correlation-based learning models that use Hebbian learning over large batches of inputs (Linsker, 1986; Miller, 1994), self-organizing map models operate using incremental Hebbian learning rules.”
The emergence of contrast-invariant orientation tuning in simple cells of cat visual cortex. (2007, background): “Membranepotential responses are therefore perfectly contrast invariant (Figure 1K), and the power law preserves invariance in the spike responses while narrowing the tuning width at all contrasts equally (Figure 1L) (Anderson et al., 2000a; Hansel and van Vreeswijk, 2002; Miller, 1994).”
On Advances in Statistical Modeling of Natural Images (2004, background): “Use of Gabor wavelets is also motivated by the fact that the receptive fields of simple cells in the visual cortex of animals have been found to resemble Gabor functions [58].”
Modeling sleep and wakefulness in the thalamocortical system. (2005, background): “Several other computational models have explored the genesis of orientation and directional selectivity (Douglas et al. 1995; McLaughlin et al. 2000; Miller 1994; Somers et al. 1995; Wimbauer et al. 1997a,b; Worgotter and Koch 1991).”
Synaptic modification by correlated activity: Hebb's postulate revisited. (2001, methodology): “…have also been successfully used in modeling activity-driven development of simple receptive fields and the ordered arrangement of ocular dominance and orientation columns (Miller et al 1989, Miller 1994), as well as ocularly matched orientation and ocular dominance maps (Erwin & Miller 1998).”
SPIKING NEURON MODELS Single Neurons , Populations , Plasticity (2002, background): “For center-surround cells the correlation is a Mexican hat-shaped function of their distance (Miller, 1994; Wimbauer et al., 1997a), e.”
INNATENESS AND EMERGENTISM (1996, background): “And yet we know that such structures do emerge reliably every time a multilayered neural network is forced to extract three-dimensional information from a two-dimensional array (e.g. Miller, 1994; Shatz, 1996).”
Computational Maps in the Visual Cortex (2005, methodology): “Most CBL models focus on individual neurons or small groups of neurons, with the exception of those of Miller (1994), Miller et al. (1989), and Erwin and Miller (1998).”
Topographica: Building and Analyzing Map-Level Simulations from Python, C/C++, MATLAB, NEST, or NEURON Components (2008, unknown): “For instance, models that represent inputs not as individual patterns but as correlation functions (e.g. Miller, 1994) are difﬁ cult to connect to Topographica, because most of the functionality of Topographica requires testing the response to speciﬁ c external stimuli (e.g. for measuring maps,…”
Building a mechanistic model of the development and function of the primary visual cortex. (2012, unknown): “Subtractive normalization with upper and lower bounds could be used instead, but it would lead to binary weights [50] [51], which is not desirable for a firing-rate model whose connections represent averages over multiple physical connections.”
Neural mechanisms of orientation selectivity in the visual cortex. (2000, background): “Miller (1994) showed that the arrangement of ON and OFF inputs to a simple cell predicted by the feed-forward model can develop from activitydependent rules of synaptic modification, provided that spontaneous LGN activity patterns in the developing animal have certain simple structures.”
Semilinear Predictability Minimization Produces Well-Known Feature Detectors (1996, background): “Redundancy reduction is widely regarded as an important goal of unsupervised learning (see, e.g., Barlow et al. 1989; Atick et al. 1992; Schmidhuber 1994; compare also Linsker 1988; Foldiak 1990; Deco and Obradovic 1996; Miller 1994; Field 1994).”