Neural Dynamics Model of Visual Decision-Making: Learning from Human Experts

TL;DR

This work introduces a neural dynamics model of visual motion decision-making that emulates dorsal-stream processing (LGN→V1→MT→LIP) with leaky integrate-and-fire neurons and biomimetic synapses, achieving primate-like neural and behavioral patterns in the RDK task. A key advance is neuroimaging-informed fine-tuning, which leverages structural and functional MRI features (e.g., FA/white-matter volumes and resting-state FC) to optimally adjust inter-area connections and conductances, yielding improved performance with a more biologically plausible architecture. Compared with CNN baselines, the model matches human behavior closely and demonstrates greater robustness to perturbation, while requiring fewer parameters and offering higher interpretability through attractor dynamics and top-down modulation. The results argue for incorporating neuroimaging priors into brain-inspired AI to enhance efficiency, resilience, and explainability, informing future development of robust, biologically grounded perceptual systems.

Abstract

Uncovering the fundamental neural correlates of biological intelligence, developing mathematical models, and conducting computational simulations are critical for advancing new paradigms in artificial intelligence (AI). In this study, we implemented a comprehensive visual decision-making model that spans from visual input to behavioral output, using a neural dynamics modeling approach. Drawing inspiration from the key components of the dorsal visual pathway in primates, our model not only aligns closely with human behavior but also reflects neural activities in primates, and achieving accuracy comparable to convolutional neural networks (CNNs). Moreover, magnetic resonance imaging (MRI) identified key neuroimaging features such as structural connections and functional connectivity that are associated with performance in perceptual decision-making tasks. A neuroimaging-informed fine-tuning approach was introduced and applied to the model, leading to performance improvements that paralleled the behavioral variations observed among subjects. Compared to classical deep learning models, our model more accurately replicates the behavioral performance of biological intelligence, relying on the structural characteristics of biological neural networks rather than extensive training data, and demonstrating enhanced resilience to perturbation.

Figures (9)

• Figure 1: A neural dynamics model of motion perception inspired by the dorsal visual pathway.a. The Random Dot Kinematogram (RDK) and the locations of the key structures within the dorsal visual pathway associated with motion perception and decision-making in the human brain. b. Spatial characteristics and arrangements of LGN ON and OFF neurons, including their spatial and temporal profiles. c. Structure of the biologically inspired neural dynamics model, illustrating four modules corresponding to those key brain regions and the connections between them. d-f. Averaged firing rates of distinct groups of neurons in response to the RDK stimulus. Time 0 marks the stimulus onset. The solid line represents neuron groups with a preferred directional bias, while the dotted line represents neuron groups with an opposite bias. Color and legend indicate coherence levels. d. V1 neurons demonstrate direction selectivity; e. MT neurons enhance this selectivity, with the activation of direction-preference neurons increasing with coherence; f. LIP neurons exhibit the ramping activity and a winner-take-all effect, with the slope of ramping increasing as coherence levels rise.
• Figure 2: Behavioral performance of the model in RDK tasks and effect of virtual electrical stimulation.a. Behavioral performance of the model and human subjects in the RDK task, showing selection probability, and average reaction time as functions of coherence level (upper: psychometric curve, lower: decision time curve). The model's performance is represented in blue (average of 45 repetitions), and human performance is shown in orange (average of 36 subjects). Error bars denote the standard error of the mean (SEM) across all subjects. The inset subplot displays the distribution of sensitivity ($k$) for all subjects (orange) and the model (blue, repeated 45 times). b. When additional current (040) is injected into the 'L' group of neurons in the MT area of the model, the psychometric and decision time curves shift rightward, sensitivity decreases (slope=-0.25, $p<0.001$), and reaction time shorten. Color and colorbar represent the intensity of stimulation, consistent throughout. c. Injecting additional current into the 'R' group of neurons in MT results in a leftward shift in the psychometric and decision time curves, with decreased sensitivity and decision time (slope=-0.26, $p<0.001$). d. Performance changes when additional electrical stimulation is applied to all neurons in the model's V1. The additional input current decreases the slope of the psychometric curve (indicating reduced sensitivity, slope=-0.50, $p<0.001$) and reduces mean decision time. e. Sensitivity and decision time both decrease when additional electrical stimulation is applied to all neurons in MT (slope=-0.26, $p<0.001$). f. Applying the same intensity of current to LIP has negligible effects on the model's performance (slope=0.005, $p=0.60$).
• Figure 3: Neuroimaging-informed model tuning.a. Mean FA values of the left lateral occipital area in V1 were negatively correlated with subjects' task performance. The red dot represents the best-performing subject (hereinafter the same). Subjects who exhibited lower anisotropy (higher fiber bundle density) in this region performed better in the task. b. Tuning the average connection weight between V1 and MT in the model improved accuracy and efficiency within a specific range. Upper: Schematic diagram of the adjusted V1--MT connection. Lower: Increasing the connection weight initially improves model sensitivity and reduces decision time, but later stages show a decrease in sensitivity while decision time continues to decrease. c. The white matter volume in the right inferior parietal lobe, located between the occipital and parietal lobes, was positively correlated with task performance. This indicates that subjects with better performance had larger white matter volume relative to total brain volume. d. Tuning the proportion of connections between MT and LIP within a specific range improved the model's accuracy and efficiency. Upper: Schematic diagram of the adjusted MT--LIP connection. Lower: Increasing the connection ratio initially enhances sensitivity and reduces decision time, but excessive connections can decrease sensitivity. e. Resting-state functional connectivity between MT and AAIC in the right hemisphere was positively correlated with task performance. The positive connectivity may reflect enhanced self-monitoring during the task, leading to improved performance. f. Increasing the average synaptic conductivity in MT within a specific range improved the model's sensitivity and efficiency. Upper: Schematic diagram of the adjusted MT--LIP connections. Lower: As conductivity increases, sensitivity initially rises and decision time decreases, but sensitivity eventually declines despite continued reduction in decision time. g. Functional connectivity between LIP and the suborbital region in the left hemisphere was negatively correlated with task performance. The negative connectivity may reflect increased engagement during the task, which could lead to improved performance. h. Increasing the Hebb-strengthened weight between LIP excitatory neurons within a specific range improved the model's sensitivity and efficiency. Upper: Schematic diagram of the adjusted LIP recurrent connections. Lower: Increasing the weight initially enhances sensitivity and reduces decision time, but excessive weight can decrease sensitivity. The Pearson correlation (uncorrected) was used for these analyses; $p$-values are reported.
• Figure 4: Comparison between CNN and the neural dynamics model in terms of perturbation.a. The structure and parameters of the CNN model. b-e. In the CNN model, changes in accuracy under different perturbation conditions: b. when a certain percentage of connections are dropped in each layer; c. when zero-mean Gaussian noise (with variance as a multiplier of the average absolute value) is added to the connection weights; d. when a certain percentage of neurons are dropped; and e. when zero-mean Gaussian noise is added to the neuron activations. The colors represent the various parameter layers of the network (b, c, excluding the fully connected layer) or the presynaptic neurons (d, e). Cross marks indicate the accuracy of individual experiments, while the curves represent the overall accuracy averaged across five experiments. f-i. In the neural dynamics model, changes in accuracy under similar perturbation conditions: f. when a certain percentage of connections are dropped in each layer; g. when zero-mean Gaussian noise is added to the connection weights; h. when a certain percentage of neurons are dropped; and i. when zero-mean Gaussian noise is added to the neuron activations. The colors represent the neurons in each layer (h, i) or the connections emanating from each layer (f, g).
• Figure 5: Diagram showing the impact of parameter adjustments on the model.a. Adjusting connections between V1 and MT or MT and LIP alters the information received by LIP. Increasing connection weights shifts the input currents received by the two groups of direction-selective neurons in LIP from state 1 to state 2, thereby increasing both the differences in input currents between the two groups and their absolute magnitudes. b. Changes in the information received by LIP affect the energy landscape and attractors in the decision space. Insufficient input currents fail to trigger a decision, maintaining LIP neurons at state 1; optimal input currents and current differences drive the state transition to decision state 2; excessively large input currents cause the state to shift towards 3, deviating from the intended decision state. c. Changes in LIP's recurrent connections also alter the landscape and attractor states in the decision space. Insufficient recurrent connections result in weak evidence being unable to drive LIP from resting state 1 to decision state 2; excessive recurrent connections cause the resting attractor state 1 to disappear and enlarge the attraction basins of decision states 2 and 3, where weak inputs or noise can drive LIP to one of these decision states.