Unlike most artificial systems, the brain is able to face situations that it has not learned or even encountered before. This ability is not in general echoed by the properties of most neural networks. Here, we show that neural computation based on least-square error learning between populations of intensitycoded neurons can explain interpolation and extrapolation capacities of the nervous system in sensorimotor and cognitive tasks. We present simulations for function learning experiments, auditory-visual behavior, and visuomotor transformations. The results suggest that induction in human behavior, be it sensorimotor or cognitive, could arise from a common neural associative mechanism.

Working memory performance is modulated by the level of dopamine (DA) D1 receptors stimulation in the prefrontal cortex (PFC). This modulation is exerted at different time scales. Injection of D1 agonists/antagonists exerts a long-lasting influence (several minutes or hours) on PFC pyramidal neurons. In contrast, during performance of a cognitive task, the duration of the postsynaptic effect of phasic DA release is short lasting. The functional relationships of these two time scales of DA modulation remain poorly understood. Here we propose a model that combines these two time scales of DA modulation on a prefrontal neural network. The model links the cellular and behavioral levels during performance of the delayed alternation task. The network, which represents the activity of deep-layer pyramidal neurons with intrinsic neuronal properties, exhibits two stable states of activity that can be switched on and off by excitatory inputs from long-distance cortical areas arriving in superficial layers. These stable states allow PFC neurons to maintain representations during the delay period. The role of an increase of DA receptors stimulation is to restrict inputs arriving on the prefrontal network. The model explains how the level of working memory performance follows an inverted U-shape with an increased stimulation of DA D1 receptors. The model predicts that (1) D1 receptor agonists increase perseverations, (2) D1 antagonists increase distractability, and (3) the duration of the postsynaptic effect of phasic DA release in the PFC is adjusted to the delay period of the task. These results show how the precise duration of the postsynaptic effect of phasic DA release influences behavioral performance during a simple cognitive task.

Sensorimotor systems face complex and frequent discrepancies among spatial modalities, for example, growth, optical distortion, and telemanipulation. Adaptive mechanisms must act continuously to restore perceptual-motor alignments necessary for perception of a coherent world. Experimental manipulations that exposed participants to localized discrepancies showed that adaptation is revealed by the acquisition of a constrained relation between entire modalities rather than associations between individual exemplars within these modalities. The computational problem faced by the human nervous system can thus be conceived as having to induce constrained relations between continuous stimulus and response dimensions from ambiguous or incomplete training sets, that is, performing interpolation and extrapolation. How biological neuronal networks solve this problem is unknown. Here we show that neural processing based on linear collective computation and least-square (LS) error learning in populations of frequency-coded neurons (i.e., whose discharge varies in a monotonic fashion with a parameter) has built-in interpolation and extrapolation capacities. This model can account for the properties of perceptual-motor adaptations in sensorimotor systems.