Robert B. Pinter

Analysis and analog implementation of directionally sensitive shunting inhibitory neural networks

The most significant property of these networks is their differential response to stimuli moving in opposite directions. A quantitative analysis shows that this directional response adapts to mean luminance levels and varies with size and speed of moving objects, as well as with coupling order among elements of a network. Both biophysical and analog hardware implementations of this class of networks are given here. Implementation of unidirectional coupling and the response to directional edges are demonstrated and shown to accord well with that of the neural network.

Frequency and time domain properties of retinular cells of the desert locust (Schistocerca gregaria) and the house cricket (Acheta domesticus)

Summary1.Light adapted retinular cell responses of the desert locust (Schistocerca gregaria) are shown to have finer time resolution and higher frequency response than those of the house cricket (Acheta domesticus), and these responses conform to earlier observations relating flicker-fusion frequencies of the electroretinogram to flight behavior in certain insects.2.The amplitude and phase frequency response characteristics and their dependence on mean intensity for the locust and the cricket are similar toLimulus, but distinctly different fromCalliphora (Zettler, 1969) at similar light intensities. The relationship of the Fuortes-Hodgkin model to these observations is discussed.3.The signal to noise ratios of the sinusoidal responses are never higher than 4 to 1.4.Pulse responses of both the locust and the cricket are predicted from the inverse Fourier transform of the frequency response.5.The three measured parameters in the response to steps of light, i.e., latency, initial slope and final level in the steady state, and incremental gain as a function of mean light intensity are similar for the locust and the cricket andLimulus, and the Fuortes-Hodgkin model.

Primitive features by steering, quadrature, and scale

The impulse response of neurons in the visual cortex of the mammalian brain has been known for some time. How to make use of these as filters has led to many hypotheses. The response of a single filter is ambiguous because the result depends on stimulus type, contrast, position, orientation, and scale. We show that a set of quadrature filters at sparse positions can be constructed so that it is possible to disambiguate the 2D responses of the individual filters. Detecting edges is not the goal of the present work; rather, we seek to detect relevant edges. Thus, we make the assumption that at the scale of interest, a local image patch consists predominantly of an edge or a bar. When this patch is processed by five or seven oriented filters, one can compute the exact orientation and centroid position of the feature. When the set of filters is applied at two different scales, it is possible to distinguish edges from ridges and to identify the polarity, intensity, and width. It is also possible to find corners and blobs. These computations are stable under image shifts in position and orientation and can be made to subpixel resolution.

Adaptation of spatial modulation transfer functions via nonlinear lateral inhibition

Adaptation, or change of shape of spatial modulation transfer functions (MTFs) on change of mean luminance level, occurs in visual interneurone and human psychophysical observations. Generally the bandwidth decreases and relative low frequency attenuation decreases as mean luminance decreases. Here it is shown how these changes in MTFs can be accounted for by nonlinear lateral inhibition based on spatial distributions of efficacy of voltage controlled synaptic conductance variation.

Adaptation of receptive field spatial organization via multiplicative lateral inhibition

The interactions among electrically independent neurons via synapses mediating voltage controlled conductance become primarily multiplicative lateral inhibition. This nonlinear lateral inhibition among members of an array of neurons causes adaptation of the organization of the spatial receptive field. A proof of the adaptation is given and applications of the results to studies on insect visual interneurons are discussed. Given a simple hypothesis of spatial gradient of order of conductance dependence on neighboring cell voltage, a sharpening of spatial tuning of the receptive field is predicted with increased background level along with an increased linearization of the neuronal response function.

The electrophysiological bases for linear and for nonlinear product term lateral inhibition and the consequences for wide field textured stimuli.

The electrophysiological bases of linear, and of nonlinear product term recurrent lateral inhibition are defined and the general equations derived. Defining texture as sinusoidal spatial luminance functions, the response characteristics of nonlinear product term lateral inhibitory arrays to wide field textured stimuli are derived, and applications to locust DCMD and Y retinal ganglion cells discussed.

Product term nonlinear lateral inhibition enhances visual selectivity for small objects or edges

Abstract Hartline-Ratliff recurrent linear lateral inhibition enhances contrast in a visual system but does not account for severely nonlinear small object selectivity often found in insect movement detector systems. It is shown here that product term recurrent nonlinear lateral inhibition, homologous to voltage sensitive conductance, can mediate the small object sensitivity.

Original Contribution: Implementation of front-end processor neural networks

Electronic implementation of a class of neural networks whose short-term memory equation is governed by multiplicative, rather than additive, inhibition is proposed. The network models can be derived from ionic flow in nerve membranes and multiplicative terms result from control of conductive paths by voltages of other cells in the network. Since Field Effect Transistors (FETs) are voltage controlled conductances when operated below pinch-off, these networks can be readily implemented in FET technology using this physical property. This class of neural networks appears in many areas of the brain as well as the sensory system and has been used as a basic building block for the multilayer self-organizing architecture of Adaptive Resonance Theory (ART). The model has been especially useful for explaining a wide range of peripheral visual phenomena. The implementation is intended to specifically demonstrate desirable front-end image processing properties of contrast enhancement, edge detection, dynamic range compression, and adaptation of dynamics to mean intensity levels. Since the network can be mathematically described, its dynamics and stability may be examined. Compatibility of the network with higher level processing allows for its inclusion in multilayer self-organizing neural network architectures.

A shunting inhibitory motion detector that can account for the functional characteristics of fly motion-sensitive interneurons

The multiplicative inhibitory motion detector (MIMD) has been proposed by the authors (Visual Communications and Image Processing IV, Proceedings of SPIE. 1199, 1229-1240, 1989). The applicability of this motion detector to the activity of motion-sensitive interneurons of the lobula plate, the posterior pat of the third visual ganglion in the flys optic lobe, is investigated. In particular, it is demonstrated that an array of MIMDs can simulate the characteristics of transient and steady-state response and of contrast sensitivity functions for these neurons

Shunting neural network photodetector arrays in analog CMOS

This paper describes a custom analog CMOS photodetector array IC that exploits nonlinear lateral inhibition to achieve dynamic range compression, edge enhancement, and adaptation to mean input intensity. The neural net array architecture, characterized by nearest-neighbor connections and multiplicative cell interaction, is modeled after biological vision systems. The fabricated IC successfully implements a portion of the compact and powerful nonlinear signal processing performed in the outer layers of the vertebrate retina. Measured results are presented for an optical input intensity range of nearly six decades. A scanning architecture that allows for preferential directional sensitivity is also demonstrated. Measured data agree well with models created using a spreadsheet program. >