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

Abstract

Perceptual decision-making is often modeled as the accumulation of sensory evidence over time. Recent studies using psychophysical reverse correlation have shown that even though the sensory evidence is stationary over time, subjects may exhibit a time-varying weighting strategy, weighting some stimulus epochs more heavily than others. While previous work has explained time-varying weighting as a consequence of static decision mechanisms (e.g., decision bound or leak), here we show that time-varying weighting can reflect strategic adaptation to stimulus statistics, and thus can readily take a number of forms. We characterized the temporal weighting strategies of humans and macaques performing a motion discrimination task in which the amount of information carried by the motion stimulus was manipulated over time. Both species could adapt their temporal weighting strategy to match the time-varying statistics of the sensory stimulus. When early stimulus epochs had higher mean motion strength than late, subjects adopted a pronounced early weighting strategy, where early information was weighted more heavily in guiding perceptual decisions. When the mean motion strength was greater in later stimulus epochs, in contrast, subjects shifted to a marked late weighting strategy. These results demonstrate that perceptual decisions involve a temporally flexible weighting process in both humans and monkeys, and introduce a paradigm with which to manipulate sensory weighting in decision-making tasks.

Timescales of Evidence Evaluation for Decision Making and Associated Confidence Judgments Are Adapted to Task Demands

“…Previous studies have shown that a variety of species adapt their temporal weighting of evidence based on task demands. Humans and monkeys adopt temporal weighting of evidence to match learned temporal regularities of stimuli that are repeatedly presented, giving more weight to time periods with higher fidelity information (Levi et al, 2018). In that study, visual stimuli were presented for a fixed amount of time with no underlying variability in the time at which the information fidelity changed.…”

“…Shorter timescales are required for detecting changes in more volatile environments and when signal durations are briefer. Humans have been shown to adapt their timescale of evidence evaluation accordingly as a function of the distribution of signal durations and timing for visual stimuli (Tsetsos et al, 2012;Ossmy et al, 2013;Bronfman et al, 2016;Levi et al, 2018). Humans have also been shown to adapt their timescales of evidence evaluation in the optimal direction as a function of volatility when classifying source distributions for the location of visual stimuli (Glaze et al, 2015).…”

Perceptual rivalry across animal species

“…Not surprisingly then, macaque visual performance has consistently been found to be close to that of humans (Buracas & Albright, 1999;Elmore & Wright, 2015;Gelfand & Horwitz, 2018;Harwerth, 1982;Kaas, 1992;Levi, Yates, Huk, & Katz, 2018;Lynn & Curran, 2010;Rajalingham, Schmidt, & DiCarlo, 2015;Zarco, Merchant, Prado, & Mendez, 2009). Macaques have also been shown to perceive multistable visual stimuli perceptually fluctuating with a temporal profile that is similar to that of humans (Dodd, Krug, Cumming, & F I G U R E 3 Evidence for perceptual rivalry in macaque.…”

Stimulus-choice (mis)alignment in primate MT cortex

“…Figure 1 demonstrates this alignment conceptually and the effect of 27 each type of choice model in this space. Specifically, for a population of only two 28 neurons, the joint activity of the population can be represented as points in a 2D space 29 where each axis represents an individual neuron's activity ( Figure 1A). For a 30 one-dimensional stimulus (as is typically used in discrimination paradigms), different 31 values of the stimulus (red and black) will drive activity that falls along a The choice is fed back and pushes variability along the stimulus axis.…”

“…These results suggest the 76 monkeys' readout is either highly suboptimal or choice signal is fed back to sensory 77 cortex in the null space of the stimulus -multiplexing choice signals in sensory areas 78 without corrupting information about the stimulus. This feedback signal could be 79 critical for adapting sensory representations while learning new tasks or in 80 non-stationary environments [27,28].…”