Non-monotonic Temporal-Weighting Indicates a Dynamically Modulated Evidence-Integration Mechanism

Abstract

Perceptual decisions are thought to be mediated by a mechanism of sequential sampling and integration of noisy evidence whose temporal weighting profile affects the decision quality. To examine temporal weighting, participants were presented with two brightness-fluctuating disks for 1, 2 or 3 seconds and were requested to choose the overall brighter disk at the end of each trial. By employing a signal-perturbation method, which deploys across trials a set of systematically controlled temporal dispersions of the same overall signal, we were able to quantify the participants’ temporal weighting profile. Results indicate that, for intervals of 1 or 2 sec, participants exhibit a primacy-bias. However, for longer stimuli (3-sec) the temporal weighting profile is non-monotonic, with concurrent primacy and recency, which is inconsistent with the predictions of previously suggested computational models of perceptual decision-making (drift-diffusion and Ornstein-Uhlenbeck processes). We propose a novel, dynamic variant of the leaky-competing accumulator model as a potential account for this finding, and we discuss potential neural mechanisms.

Strategic and Dynamic Temporal Weighting for Perceptual Decisions in Humans and Macaques

“…Just as past work has taken such early weighting as a signature of bounded accumulation, late weighting has been posited to reflect leaky integration. However, such models have been increasingly updated to accommodate either sort of behavioral signature (Usher and McClelland, 2001; Tsetsos et al, 2012; Bronfman et al, 2016). Thus, while time varying weighting has been identified before, it is almost always discussed as diagnostic about the structure of a decision-making mechanism, i.e., perfect or leaky integration to a bound (fixed or collapsing).…”

“…This notion is supported by experiments in which weighting changes systematically with variable trial length and signal timings (Ghose 2006; Tsetsos et al, 2012; Ossmy et al, 2013; Bronfman et al, 2016), as well as by studies that explore effects of congruency between serially presented samples (Cheadle et al, 2014). Irrespective of a stipulated model or mechanism, these studies point to similar conclusions: subjects may reweigh stimulus information as dictated by the reliability of the evidence and demands of the task.…”

“…In many instances, subjects have exhibited “early weighting,” where sensory evidence presented in early epochs contributes more to choices than that in late (Huk and Shadlen, 2005; Kiani et al, 2008, Nienborg and Cumming, 2009; Yates et al, 2017). In other instances, however, “late weighting” has been observed, where choices were primarily influenced by sensory evidence presented in late stimulus epochs (Tsetsos et al, 2012; Cheadle et al, 2014; Bronfman et al, 2016; Carland et al, 2016). In rodents, a mixture of either early or flat weighting profiles has been reported (Erlich et al, 2015; Scott et al, 2015; Pinto et al, 2017; Licata et al, 2017).…”

An accumulation-of-evidence task using visual pulses for mice navigating in virtual reality

“…In particular, it has been argued that primacy in pulsedbased tasks is due to either reaching an accumulator bound (Kiani et al, 2008) or to competition between leaky integrators that mutually inhibit each other (Tsetsos et al, 2012) . Interestingly, it has been recently shown that in humans the degree of primacy and even the monotonicity of the evidence weighting curve can change with stimulus duration, which prompted the authors to postulate a dynamic evidence accumulation mechanism (Bronfman et al, 2016) . Thus, it is conceivable that different decisionmaking and integration mechanisms might be at play depending on stimulus and task features (Piet et al, 2017; Uchida et al, 2006 .…”

Choices Change the Temporal Weighting of Decision Evidence

“…In the perceptual choice tasks commonly used in the laboratory (but see [3][4][5]), performance is maximized by weighing evidence equally across time [1]. Yet, the evidence weighting applied by human and non-human decision-makers often deviates substantially from such flat weighting profiles (but see [6,7]): some studies found stronger weighting of early evidence ('primacy'; [8][9][10][11]), others stronger weighting of late evidence ('recency'; [12][13][14]), and yet others even non-monotonic weighting profiles [15].…”