Testing the state-dependent model of subsecond time perception against experimental evidence

Coordinated movements, speech, and other actions are impossible without precise timing. Computational models of interval timing are expected to provide key insights into the underlying mechanisms of timing, which are currently largely unknown. So far, existing models have only been partially replicating key experimental observations, such as the linear psychophysical law, the linear increase of the standard deviation (the scalar property or Weber’s law), and the modulation of subjective duration via dopamine. Here, we incorporate the state-dependent model for subsecond timing as proposed by Buonomano (2000) into a strongly data-driven computational network model of PFC We show that this model variant, the state-dependent PFC model, successfully encodes time up to 750 milliseconds and reproduces all key experimental observations mentioned above, including many of its details. Investigating the underlying mechanisms, we find that the representations of different intervals are based on the natural heterogeneity in the parameters of the network, leading to stereotypic responses of subsets of neurons. Furthermore, we propose a theory for the mechanism underlying subsecond timing in this model based on correlation and ablation analyses as well as mathematical analyses explaining the emergence of the scalar property and Vierordt law. The state-dependent PFC model proposed here constitutes the first data-driven model of subsecond timing in the range of hundreds of milliseconds that has been thoroughly tested against a variety of experimental data, providing an ideal starting point for further investigations of subsecond timing.