Neural dynamics of adaptive timing and temporal discrimination during associative learning

Abstract A neural network model that controls behavioral timing is described and simulated. This model, called the Spectral Timing Model, controls a type of timing whereby an animal or robot can learn to wait for an expected goal by discounting expected nonoccurrences of a goal object until the expected time of arrival of the goal. If the goal object does not then materialize, the animal can respond to unexpected nonoccurrences of the goal with appropriate changes in information processing and exploratory behavior. The model is a variant of the gated dipole model of opponent processing. When the gated dipole model is generalized to include a spectrum of cellular response rates within a large population of cells, the model's total output signal generates accurate learned timing properties that collectively provide a good quantitative fit to animal learning data. In particular, the Spectral Timing Model utilizes the habituative transmitter gates and adaptive long-term memory traces that are characteristic of gated dipole models. The Spectral Timing Model is embedded into an Adaptive Resonance Theory (ART) neural architecture for the learning of correlations between internal representations of recognition codes and reinforcement codes. This type of learning is called conditioned reinforcer learning. The two types of internal representations are called sensory representations (S) and drive representations (D). Activation of a drive representation D by the Spectral Timing Model inhibits output signals from the orienting subsystem (A) of the ART architecture and activates a motor response. The inhibitory pathway helps to prevent spurious resets of short-term memory, forgetting, and orienting responses from being caused by events other than the goal object prior to the expected arrival time of the goal. Simulated data properties include the inverted U in learning as a function of the interstimulus interval (ISI) that occurs between onset of the conditioned stimulus (CS) and the unconditioned stimulus (US); correlations of peak time, standard deviation, Weber fraction, and peak amplitude of the conditioned response as a function of the ISI; increase of conditioned response amplitude, but not its timing, with US intensity; speed-up of the timing circuit by an increase in CS intensity or by drugs that increase concentrations of brain dopamine or acetylcholine; multiple timing peaks in response to learning conditions using multiple ISIs; and conditioned timing of cell activation within the hippocampus and of the contingent negative variation (CNV) event-related potential. The results on speed-up by drugs that increase brain concentrations of dopamine and acetylcholine support a 1972 prediction that the gated dipole habituative transmitter is a catecholamine and its long-term memory trace transmitter is acetylcholine. It is noted that the timing circuit described herein is only one of several functionally distinct neural circuits for governing different types of timed behavior competence.