Interval timing by neural integration: Supplementary Materials (original) (raw)

In these Supplementary Materials, we describe additional neural mechanisms needed to justify certain details of the timing model described in the main text, and we supplement the human behavioral evidence for rapid duration learning with evidence from mice performing a timing task. As discussed in section 1, the additional neural mechanisms include a network that performs temporal di erentiation to decode the time left before an upcoming event, and a feedback controller that uses di erentiation to stabilize the integrator dynamics of the timing model. In section 2, we derive the linear noise assumption used in the main text from the sigmoidal activation function of our leaky integrator model of neural population activity. Section 3 demonstrates that the presence of a lower reflecting boundary in a drift-di usion model of neural activity still leads to an approximately inverse Gaussian distribution of response times (RTs), despite the strict dependence of this inverse Gaussian shape ...