Spatiotemporal dynamics of random stimuli account for trial-to-trial variability in perceptual decision making

Abstract

Decisions in everyday life are prone to error. Standard models typically assume that errors during perceptual decisions are due to noise. However, it is unclear how noise in the sensory input affects the decision. Here we show that there are experimental tasks for which one can analyse the exact spatio-temporal details of a dynamic sensory noise and better understand variability in human perceptual decisions. Using a new experimental visual tracking task and a novel Bayesian decision making model, we found that the spatio-temporal noise fluctuations in the input of single trials explain a significant part of the observed responses. Our results show that modelling the precise internal representations of human participants helps predict when perceptual decisions go wrong. Furthermore, by modelling precisely the stimuli at the single-trial level, we were able to identify the underlying mechanism of perceptual decision making in more detail than standard models.

A Bayesian Reformulation of the Extended Drift-Diffusion Model in Perceptual Decision Making

“…In accordance with previous results (Park et al, 2016), group-level model comparison (Figure 7) indicated that the exact input models (EXaM and eEXaM) explain choice and response times better than the DDM-equivalent models without such exact input (BM and eBM). Furthermore, the fitted values for the “bound” parameter were significantly higher in the exact input models compared to the DDM-equivalent models (Tables 2, 3).…”

“…Coupling the internal uncertainty ${\hat{\sigma }}^b$ to the noise level σ b through Equation (6) deviates from our previous approaches of letting ${\hat{\sigma }}^b$ vary freely (Bitzer et al, 2014), or fixing it to a constant, stimulus-derived value (Park et al, 2016). Coupling ${\hat{\sigma }}^b$ to σ b is necessary in the eBM to implement the effect of slow errors, that is, the effect that errors are slower than correct responses (cf.…”

“…In the BM we also use 2 effective parameters: noise level (σ b ) and bound while fixing the internal uncertainty (${\hat{\sigma }}^b$), to a value derived from σ b (Equation 6). We included the remaining parameters (related to lapses, non-decision time and bias), because we have found in previous work that they improve fits of the model to behavioral data (Park et al, 2016). …”

“…Section Internal Uncertainty and Arbitrary Scaling of Model Parameters). Additionally, we have previously introduced three extra parameters: variability of non-decision time, $s_t^b$, lapse probability, ${\pi }_l^b$, for modeling entirely random lapses, and time out lapse probability, ${\pi }_{to}^b$ (Park et al, 2016; see Section Methods for more details).…”

“…This means that one can model the trial-wise stimulus and improve model fits profoundly (Park et al, 2016). …”