Interval timing as optimal inference: Investigating aspects of timing behaviours from the perspective of Bayesian inference

How time intervals are maintained in and retrieved from memory is a current issue of debate. Recently it has been suggested that interval timing follows a Bayesian inference process whereby people incorporate the perceived interval and their prior experience (e.g., previous experienced durations) to make an optimal “guess” (e.g., Jazayeri & Shadlen, 2010). While the Bayesian approach accounts well for observations in time reproduction, whether and how it can be applied to other timing behaviours is yet unclear. In addition, the detailed mechanism and neural substrates of Bayesian timing still remain to be explored. This project will investigate these aspects of timing behaviours from the perspective of Bayesian inference.

First, it will investigate whether and how the Bayesian approach can be extended to model a wide range of timing behaviours, including timing biases in time reproduction (e.g., under-reproduction; Riemer et al., 2012), time comparison (e.g., shortened bisection point, time-order error; Kopec & Brody, 2010; Hellström & Rammsayer, 2015) and the cross-dimensional interaction between time and other physical dimensions (e.g., space; Cai & Connell, 2015, 2016). We aim to build computational models to simulate human behaviours in these different timing tasks.

Second, the project aims to develop a mechanistic account of Bayesian timing. It has been argued that, when estimating how much time has been perceived, people make an inference based on their prior belief and the evidence they received (i.e., the stimulus duration). While previous research has assumes that the prior is based on previously experienced durations and the inference is an “updated” version of the encoded/stimulus duration (Jazayeri & Shadlen, 2010), these remain hypotheses that await further empirical investigation. Such research work will form the foundation for further work on questions such as whether time is represented in memory in a modality-specific or –general fashion (e.g., Kanai et al., 2011) and whether people can maintain multiple priors and make multiple inferences at the same time.

Third, the project will also look at the neural substrates underlying Bayesian timing. In particular, it is currently unclear what brain network is involved in forming the prior from previous timing experience and whether the prior is represented in the brain according to the modality of previous durations. To do this, we will use brain stimulation (e.g., TMS or tDCS) to create virtual lesions and examine their effect on the use of prior experience in Bayesian timing.

Interested applicants can work on one or more of the above lines of work and are encouraged to engender their own research ideas related to Bayesian timing.