Humans and animals infer temporal regularities, anticipate events, and plan their actions in time. The degree to which we can exploit temporal contingencies depends on the fidelity of time measurements. Decades of behavioral research has shown that time measurements are variable and this variability scales with elapsed time. Mounting behavioral evidence demonstrates that humans and animals utilize prior information and multiple measurements to reduce the inherent uncertainty in sensory measurements. However, the computations required to reduce uncertainty in the face of scalar variability remain largely unexplored. Here, I use a Bayesian model to demonstrate that, unlike classic sensory cue combination paradigms, the computations required to combine noisy time-interval measurements optimally are nonlinear. The predictions of the optimal model are then tested against the behavior of humans performing a time-interval reproduction task. Results indicate that human performance nearly matches the predictions of the optimal model and exceed the performance of simpler linear models, demonstrating that the underlying neural computations can support nonlinear integration. 

To understand how this computation is implemented by the brain, I recorded neural activity in the medial frontal cortex (MFC) of a nonhuman primate performing a similar time-interval reproduction task. Firing rate dynamics of individual neurons in MFC were complex and heterogeneous, and were predictive of the animal’s behavior. To gain insight into how the MFC activity might represent the animal’s ongoing estimate of the of elapsed time, I analyzed the evolution of activity across the population as a function of time. Interestingly, the pattern of the population response was similar across interval durations, but the profile of activation was stretched or compressed in time to match the animal’s expectation of the interval duration. These results suggest that a key variable linking the neural dynamics to the nonlinear computations inferred from modeling the animal’s timing behavior is the speed with which the neural responses evolve.