Using the past to estimate sensory uncertainty

Beierholm, Ulrik; Rohe, Tim; Ferrari, Ambra; Stegle, Oliver; Noppeney, Uta

doi:10.7554/elife.54172

Cited by 25 publications

(29 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To form a more reliable percept of the environment, the brain needs to estimate sensory uncertainty (Beierholm et al, 2020). For instance, the magnitude of the CTB is related to the magnitude of uncertainty in sensory measurement and the select cost function (Mamassian & Landy, 2010).…”

Section: The Precision Of the Likelihood And Prior Distributionsmentioning

confidence: 99%

The temporal context in bayesian models of interval timing: Recent advances and future directions.

Sadibolova¹,

Terhune²

2022

Behavioral Neuroscience

View full text Add to dashboard Cite

Sensory perception, motor control, and cognition necessitate reliable timing in the range of milliseconds to seconds, which implies the existence of a highly accurate timing system. Yet, partly owing to the fact that temporal processing is modulated by contextual factors, perceived time is not isomorphic to physical time. Temporal estimates exhibit regression to the mean of an interval distribution (global context) and are also affected by preceding trials (local context). Recent Bayesian models of interval timing have provided important insights regarding these observations, but questions remain as to how exposure to past intervals shapes perceived time. In this article, we provide a brief overview of Bayesian models of interval timing and their contribution to current understanding of context effects. We then proceed to highlight recent developments in the field concerning precision weighting of Bayesian evidence in both healthy timing and disease and the neurophysiological and neurochemical signatures of timing prediction errors. We further aim to bring attention to current outstanding questions for Bayesian models of interval timing, such as the likelihood conceptualization.

show abstract

Section: The Precision Of the Likelihood And Prior Distributionsmentioning

confidence: 99%

The temporal context in bayesian models of interval timing: Recent advances and future directions.

Sadibolova¹,

Terhune²

2022

Behavioral Neuroscience

View full text Add to dashboard Cite

show abstract

“…As shown by Bayesian models, the interaction of two multisensory inputs can be decomposed into two contributions: one that depends on the physical discrepancy between the cues, and an overall binding tendency that is independent of the current stimulus (Beierholm et al, 2020; Körding et al, 2007; Odegaard et al, 2015; Park et al, 2021; Wozny et al, 2010). In typical laboratory studies, this overall binding tendency is often considered to be fixed for a given observer, reflecting their belief that the presented multisensory cues are causally related given the context of a specific experimental setting.…”

Section: Introductionmentioning

confidence: 99%

The context of experienced sensory discrepancies shapes multisensory integration and recalibration differently

Park

Kayser

2021

Preprint

View full text Add to dashboard Cite

Whether two sensory cues interact during perceptual judgments depends on their immediate properties, but as suggested by Bayesian models, also on the observer's a priori belief that these originate from a common source. While in many experiments this a priori belief is considered fixed, in real life it must adapt to the momentary context or environment. To understand the adaptive nature of human multisensory perception we investigated the context-sensitivity of spatial judgements in a ventriloquism paradigm. We exposed observers to audio-visual stimuli whose discrepancy either varied over a wider (±46°) or a narrower range (±26°) and hypothesized that exposure to a wider range of discrepancies would facilitate multisensory binding by increasing participants a priori belief about a common source for a given discrepancy. Our data support this hypothesis by revealing an enhanced integration (ventriloquism) bias in the wider context, which was echoed in Bayesian causal inference models fit to participants' data, which assigned a stronger a priori integration tendency during the wider context. Interestingly, the immediate ventriloquism aftereffect, a multisensory response bias obtained following a multisensory test trial, was not affected by the contextual manipulation, although participants' confidence in their spatial judgments differed between contexts for both integration and recalibration trials. These results highlight the context-sensitivity of multisensory binding and suggest that the immediate ventriloquism aftereffect is not a purely sensory-level consequence of the multisensory integration process.

show abstract

“…For example, the integration bias scales with the relative sensory reliability of the probed modality, in line with the notion of Bayesian optimal reliability weighted cue combination (Ernst and Bulthoff, 2004; Angelaki et al, 2009). A recent study has shown that the brain keeps track of the individual unisensory reliabilities and discounts previous experience in an exponential manner (Beierholm et al, 2020). Hence, our estimates of unisensory reliabilities are shaped by the collective past, which renders the weights assigned to individual sensory inputs during integration history dependent.…”

Section: Resultsmentioning

confidence: 99%

Cumulative multisensory discrepancies shape the ventriloquism aftereffect but not the ventriloquism bias

Kayser

Park

Heuer

2022

Preprint

View full text Add to dashboard Cite

Multisensory integration and recalibration are two processes by which perception deals with discrepant multisensory signals. In the lab, integration can be probed via the presentation of simultaneous spatially discrepant audio-visual stimuli, i.e. in a spatial ventriloquism paradigm. Recalibration here manifests as an aftereffect bias in unisensory judgements following immediate or long-term exposure to discrepant audio-visual stimuli. Despite many studies exploring the ventriloquism effect (VE) and the immediate aftereffect (VAE), it remains unclear whether the VAE is a direct consequence of the integration of preceding discrepant multisensory signals. We here provide evidence that these two biases are not strictly related. First, we analysed data from ten experiments probing the dependence of each bias on audio-visual discrepancies experienced in multiple preceding trials. This revealed a seven-fold stronger dependence of the VAE on preceding discrepancies compared to the VE. Second, we analysed data from an experiment deviating from the typical trial context in which these biases are probed and found that the VAE can vanish despite the VE being present. We argue that integration may help maintaining a stable percept by reducing immediate sensory discrepancies, whereas recalibration may help maintaining an accurate percept by accounting for consistent discrepancies. Hence, the immediate VAE is not a direct and necessary consequence of the integration of discrepant signals and these two well-studied multisensory response biases can be experimentally dissociated.

show abstract

Using the past to estimate sensory uncertainty

Cited by 25 publications

References 43 publications

The temporal context in bayesian models of interval timing: Recent advances and future directions.

The temporal context in bayesian models of interval timing: Recent advances and future directions.

The context of experienced sensory discrepancies shapes multisensory integration and recalibration differently

Cumulative multisensory discrepancies shape the ventriloquism aftereffect but not the ventriloquism bias

Contact Info

Product

Resources

About