Temporal integration is a robust feature of perceptual decisions
Curation statements for this article:-
Curated by eLife
eLife assessment
This manuscript tests an important assumption about how sensory information is processed and used to guide motor choices. The widely held assumption is that sensory-motor circuits are capable of integrating evidence, but the validity and generality of this 'principle' have been recently questioned by studies suggesting that other computational operations may lead to similar psychophysical results, mimicking integration without actually performing it. This study makes a compelling case that the integration assumption was likely correct all along and that the model mimicry can be easily disambiguated by using appropriate sensory stimuli and task designs that permit rigorous analyses.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Neuroscience (eLife)
Abstract
Making informed decisions in noisy environments requires integrating sensory information over time. However, recent work has suggested that it may be difficult to determine whether an animal’s decision-making strategy relies on evidence integration or not. In particular, strategies based on extrema-detection or random snapshots of the evidence stream may be difficult or even impossible to distinguish from classic evidence integration. Moreover, such non-integration strategies might be surprisingly common in experiments that aimed to study decisions based on integration. To determine whether temporal integration is central to perceptual decision-making, we developed a new model-based approach for comparing temporal integration against alternative ‘non-integration’ strategies for tasks in which the sensory signal is composed of discrete stimulus samples. We applied these methods to behavioral data from monkeys, rats, and humans performing a variety of sensory decision-making tasks. In all species and tasks, we found converging evidence in favor of temporal integration. First, in all observers across studies, the integration model better accounted for standard behavioral statistics such as psychometric curves and psychophysical kernels. Second, we found that sensory samples with large evidence do not contribute disproportionately to subject choices, as predicted by an extrema-detection strategy. Finally, we provide a direct confirmation of temporal integration by showing that the sum of both early and late evidence contributed to observer decisions. Overall, our results provide experimental evidence suggesting that temporal integration is an ubiquitous feature in mammalian perceptual decision-making. Our study also highlights the benefits of using experimental paradigms where the temporal stream of sensory evidence is controlled explicitly by the experimenter, and known precisely by the analyst, to characterize the temporal properties of the decision process.
Article activity feed
-
-
eLife assessment
This manuscript tests an important assumption about how sensory information is processed and used to guide motor choices. The widely held assumption is that sensory-motor circuits are capable of integrating evidence, but the validity and generality of this 'principle' have been recently questioned by studies suggesting that other computational operations may lead to similar psychophysical results, mimicking integration without actually performing it. This study makes a compelling case that the integration assumption was likely correct all along and that the model mimicry can be easily disambiguated by using appropriate sensory stimuli and task designs that permit rigorous analyses.
-
Reviewer #1 (Public Review):
This manuscript presents a comparison between models that may explain psychophysical performance in sensory integration tasks, where a subject essentially has to count stimulus samples and make a motor report about the final count.
The work has many technical strengths:
- The problem of model mimicry is clearly articulated.
- The work shows that the use of discrete sample stimulus (DSS) is key for being able to disambiguate multiple candidate mechanisms that could possibly underlie the observed behavioral data.
- The authors use rigorous model comparison and analysis techniques, some (like the integration maps) newly developed for the current application.
- The model comparison involves both qualitative and qualitative contrasts between alternative models.
- Consistent results are obtained with several data …
Reviewer #1 (Public Review):
This manuscript presents a comparison between models that may explain psychophysical performance in sensory integration tasks, where a subject essentially has to count stimulus samples and make a motor report about the final count.
The work has many technical strengths:
- The problem of model mimicry is clearly articulated.
- The work shows that the use of discrete sample stimulus (DSS) is key for being able to disambiguate multiple candidate mechanisms that could possibly underlie the observed behavioral data.
- The authors use rigorous model comparison and analysis techniques, some (like the integration maps) newly developed for the current application.
- The model comparison involves both qualitative and qualitative contrasts between alternative models.
- Consistent results are obtained with several data sets involving humans, monkeys, and rats.
- The results provide insight into why the simpler alternative models (the snapshot and extrema detection models) fail.
No glaring weaknesses were found in this manuscript. However, there are some limitations that are worth noting, to put things into context:
- The results are consistent with what has become a well-known principle of operation of sensory-motor circuits, namely, that they are highly effective at integrating sensory evidence over time. Thus, the results are not particularly surprising.
- The results are valuable in that they specifically refute two mechanisms that had been recently proposed as potential alternatives to the more standard temporal integration. To some, these alternative mechanisms may have seemed somewhat far-fetched to begin with, as they would lead to suboptimal performance in general. Nevertheless, settling the question was important.
- Temporal integration and accumulation of evidence have been the focus of many computational studies in systems neuroscience. Although these are certainly important functions, sensory-guided choices require the deployment and coordination of numerous sensory, motor, and cognitive mechanisms, of which integration is just one.
Overall, this is a valuable study that has important theoretical implications in the field of computational neuroscience. It presents a compelling case that temporal integration is a common capability of sensory-motor circuits and that it explains a variety of behavioral data sets much better than two simpler, alternative mechanisms.
-
Reviewer #2 (Public Review):
The authors' goal is to uncover the most likely method used by mammals to make choices based on a time-limited stream of noisy incoming sensory data. To achieve this, they analyze with great rigor several large datasets obtained from tightly controlled two-alternative forced choice behavioral experiments. The tight control of fluctuating incoming sensory input over a large number of trials allows the authors to extract the influence of different components of that input on the behavioral choice. The conditional analysis, showing the impact of early information on the importance of later information, or vice versa, is an excellent new technique.
They compare three models and find one based on a form of weighted integration of evidence across time is very strongly favored compared to models in which only short …
Reviewer #2 (Public Review):
The authors' goal is to uncover the most likely method used by mammals to make choices based on a time-limited stream of noisy incoming sensory data. To achieve this, they analyze with great rigor several large datasets obtained from tightly controlled two-alternative forced choice behavioral experiments. The tight control of fluctuating incoming sensory input over a large number of trials allows the authors to extract the influence of different components of that input on the behavioral choice. The conditional analysis, showing the impact of early information on the importance of later information, or vice versa, is an excellent new technique.
They compare three models and find one based on a form of weighted integration of evidence across time is very strongly favored compared to models in which only short segments of the sensory input are used, or the most extreme fluctuations of the sensory input generate a response. Overall, the results clearly do indicate that the integration-like family of models outperforms the other families. The authors succeed well in giving a fair comparison of the different families of models, allowing multiple parameters to be optimized to test different versions of each model.
It should be said that the integration model is a strange type of integration, as the weight of incoming evidence depends on the time at which it arrives-by a factor of 4 in one animal (Fig. 2)-and with an over-weighting of evidence in the middle of the sequence in one case, while the more expected effects of primacy and recency (over-weighting of early or late evidence) in another. It would be nice to see more discussion of how these differences might arise across animals, what it may say about the neural circuit performing such unbalanced integration, and how suboptimal such differential weighting of evidence is. This is important, as in some discussions integration is contrasted with state transitions, which are akin to integration over a barrier, and not necessarily ruled out by the models compared here.
-
