Transcript Bin_Rivalry
BINOCULAR RIVALRY A HIERARCHICAL MODEL FOR VISUAL COMPETETION Computational Evidence for a rivalry hierarchy in vision Wilson, PNAS (2003), Vol 100 (24), 14499-14503 Shantanu Jadhav Computational Neurobiology UCSD Outline : • What is the Binocular Rivalry – the cognitive phenomenon • Characteristics – Psychophysical features • Experimental data and evidence • The model - What it tries to explain - Implementation - Results - Predictions and limitations Lecture 1: Benefits of Computational Models - New explanations for cognitive phenomena - Tie explanations of cognitive phenomena to the biological mechanisms BINOCULAR RIVALRY • A class of phenomena characterized by fluctuating perceptual experience in the face of unvarying visual input. •Bistability as a result of ambiguous information: dissimilar images presented to the two eyes. • Competition between the two images for perceptual dominance. • Dissociation between unchanging physical stimulation and fluctuating conscious awareness => A model for studying the neural basis of conscious visual awareness. Blake and Logothetis, Nat Rev Neuro, 2002, Vol3 Perceptual Characteristics Temporal Dynamics: • Fluctuations in dominance and suppression are not regular. • No voluntary control over fluctuations • Stimulus strength, attention and visual context influence dominance periods. • Dominance and suppression rely on distinct neural processes. • Successive durations of perceptual dominance conforms to gamma distribution (universal phenomenon in bistable percepts). Spatial Features • Inter-ocular grouping during dominance => Not just suppression of an eye. (Also, figural grouping during vision rivalry) • Transitions between phases not instantaneous, but spread in a wave-like fashion Where in the visual pathway is rivalry expressed? Map NEURAL CORRELATES OF RIVALRY: EXPERIMENTAL EVIDENCE • fMRI: Modulation of activity during dominance and suppression phases in V1 (also MEGs and VERs) • Electrophysiology: No evidence for rivalry inhibition in the LGN • Modulation in Neural spiking activity in early visual cortical areas. • Increased modulation in successive stages of visual areas: MT V1 V2 V4 • Higher areas: Response only to particular preferred stimulus – stage of processing beyond the resolution of perceptual conflict. • Decrease in visual sensitivity during suppression. • Rivalry involves multiple, distributed processes throughout the rivalry hierarchy. Computational Evidence for a rivalry hierarchy in vision Wilson, PNAS (2003), Vol 100 (24), 14499-14503 • A Competitive Neural Model: Need at least two hierarchic rivalry stages for explaining data. • Specifically, the model explains the observations of a flicker and switch (F&S) procedure (which rules out inter-ocular rivalry). 18 Hz On-Off flicker of orthogonal monocular gratings + Perceptual Dominance Durations of 2.0 sec Swapping gratings between eyes at 1.5 Hz Logothetis, et al., Nature (1996), 380, 621-624 Stimulus : Right Left 0 ms 333 ms … 666 ms … 0 ms … 333 ms … 666 ms … • A single phase of perceptual dominance can span multiple alternations of the stimuli •The persistence of dominance across eye-swaps depends on temporal parameters of the stimulus • High temporal frequencies reduce the efficacy of recurrent feedback inhibition within a network • This bypasses an initial competitive inter-ocular rivalry stage, and reveals higher levels of binocular competition … IVbin IHbin EVbin EVleft EHbin EHleft IVleft IHleft EVright EHright IHright IVright Spike-Rate Equations: EVleft = Firing rate of an excitatory neuron responding to a vertical grating presented to the left eye, Asymptotic firing rate given by Naka-Rushton function EVleft drives Inhibitory Neuron Ivleft which inhibits EHright HVleft: Slow self-adaptation by an aftehyperpolarizing current Ref: Lecture 3 • Monocular Representations of horizontal and vertical gratings compete via strong reciprocal inhibition. • The competing sets of neurons self-adapt, giving rise to dominance and suppression alterations. • Spike-frequency adaptation by an Ca2+ dependent K+ current. • The second competitive stage with binocular neurons described by similar equations, with input from first layer. Vleft-bin(t) = EVleft(t) + EVright(t) • Parameters: V = 10, Emax=100, g (inhibitory gain) = 45 at monocular level, 1.53g at higher level h (hyperpolarizing current strength) = 0.47, Excitatory input gain from monocular to binocular level = 0.75 Recurrent excitation = 0.02 Stimulus = Continuous vertical grating to left eye, horizontal grating to right eye. Vertical grating response Horizontal grating response Alterations in dominance and suppression in both stages. Dominance period of 2.4 sec EHright Results: EVleft F&S stimulus Monocular Neurons cannot generate a competitive response alteration Dominance period of 2.2 sec Stronger Inhibition at binocular stage is the determining factor Conductance-based model: Simplified equations for Membrane Potential V, Recovery Variable R, inward Ca2+ current conductance T, slow Ca2+ dependent K+ hyperpolarizing conductance H Simplified equations reproduce spike shapes, firing rates and spike-frequency adaptation for human neocortical neurons Wilson HR, J. Theor. Biol. (1999), 200, 375-388 Monocular stage: 12 neurons 8 excitatory, 2 each for each eye for each grating 4 inhibitory Binocular stage: 6 neurons 4 excitatory, 2 each for each grating 2 inhibitory Parameters: TR = 4.2 msec (Exc), TR = 1.5 msec (Inh – Fast spiking cells with narrow AP) ENa = 50 mV, EK = -95mV, ECa = 120 mV, C = 1 µF, TT = 50 msec, TH = 900 msec After-hyperpolarizing current: gT = 0.1, gH =2.5 (exc) gT = 0.25, gH = 0 (inh – no spike-frequency adaptation) Conductance Model : Output of layer 1 Normal Stimulus Left Right F&S Model Gamma Distribution for Dominance Durations Variable Strength Input “A Spiking Neuron Model for Binocular Rivalry”, Laing and Chow, J. Comp. Neuro. (2002), 12, 39-53 Bifurcation Diagram for single-level Rivalry Model : g Need more inhibitory strength to produce rivalry with F&S stimulus. h Experimental and Model Results Positives : • Gamma distribution of dominance durations is obtained. • Results for F&S stimulus matched - 18.0 Hz flicker & 1.5 Hz swap by themselves give conventional rivalry • Dominance durations for variable stimulus strength reproduced. • Excitatory Feedback of max 0.02 results in similar dynamics. • Stronger inhibition at higher stages: More modulation during traditional rivalry !? • Makes clear experimental predictions Negatives : • Inter-ocular grouping not accounted for (?) • Spatial inhomogenities: Spread in a wave-like fashion. • Do we really need two layers -> for dominance durations? • Excitatory Feedback – Is it strong enough? Conclusions and Predictions Predictions • Maximum stimulus size for unitary rivalry should increase under F&S conditions. • fMRI – Blind-spot conditions : No modulation of signal during F&S. • V1 physiology: No modulation. Conclusions • Rivalry involves multiple, distributed processes throughout the visual system hierarchy • No “locus” or “neural site” of rivalry • Form vision and rivalry implemented through similar multiple networks. Grand Conclusion “Consciousness is a characteristic of extended neural circuits comprising several interacting cortical levels throughout the brain “ The Naka-Rushton Function A good fit for V1 spike rates Steady state firing rate in response to a visual stimulus of contrast P: