Gain modulation of recurrent networks
Introduction
Neuronal responses can change over short time scales due to attentional effects and processes related to motor response selection and activation. Goldberg et al. [7] have recorded neurons in area LIP that only fire to stimuli that recently have appeared in their receptive fields, or to stimuli that have behavioral significance (see also [13]). One possible mechanism for this type of change is rapid modulation of synaptic efficacy, essentially a faster form of the same processes that account for changes in selectivity over much longer time scales during learning and development [14]. A second idea is that switching arrays shift the input to the neuron being modulated [3], [10]. Here we explore another possibility, gain modulation of individual neurons within a recurrent network.
Gain modulation is a widespread mechanism by which neural responses amplitude is scaled while the selectivity of the neuron remains unchanged. Information about eye and head position is combined with visual input in parietal cortex through gain modulation of visual receptive fields [2], [4]. Gain modulation has also been seen in V4 neurons as a function of attention [5], [9]. The effects of gain modulation have been studied in feedforward networks [15], [12], [11], but not in recurrent networks. We show here that gain modulation within a recurrent circuit can dramatically affect both the activity of downstream neurons and the selectivity of the network itself.
Section snippets
Models and results
Our first model is a linear recurrent network as shown in Fig. 1a. The activity of neuron i within such a network of N neurons, ui, is determined by solvingThe first term within the parentheses is the feedforward input to neuron i, and the second term represents recurrent input from the other neurons in the network. Wij is the weight of the synapse from unit j to unit i. The parameter gi (this is a multiplicative factor not a function) is the factor by which we introduce gain
Conclusions
Modulation that changes the gain of selected neurons by a small amount can have a dramatic effect on the responses of other neurons within a recurrent network. Downstream neurons can switch between unresponsive and selectively responsive states, and network selectivity can be significantly modified. Thus, gain modulation is a good candidate mechanism for major behavioral decision functions involving switching and shaping of selectivity.
Jian Zhang is a Ph.D. student in computational neuroscience. His research explores the effects of gain modulations in cortical circuits.
References (15)
Decoding neuronal firing and modeling neural networks
Quart. Rev. Biophys.
(1994)- et al.
Encoding of spatial location by posterior parietal neurons
Science
(1985) - C.H. Anderson, D.C. Van Essen, Shifter circuits: A computational strategy for dynamic aspects of visual processing,...
- et al.
Head position signals used by parietal neurons to encode locations of visual stimuli
Nature
(1995) - et al.
Responses in area V4 depend on the spatial relationship between stimulus and attention
J. Neurophysiol.
(1996) - et al.
Recurrent excitation in neocortical circuits
Science
(1995) - et al.
The representation of visual salience in monkey parietal cortex
Nature
(1998)
Cited by (7)
Motor primitives in space and time via targeted gain modulation in cortical networks
2018, Nature NeuroscienceWhere Are the Switches on This Thing?
2009, 23 Problems in Systems Neuroscience
Jian Zhang is a Ph.D. student in computational neuroscience. His research explores the effects of gain modulations in cortical circuits.
L.F. Abbott is the Nancy Lurie Marks Professor of Neuroscience and the Director of the Volen Center for Complex Systems at Brandeis University. His research explores the effects of recurrent connectivity and both short- and long-term synaptic plasticity on the functional properties of cortical circuits.