Awareness-Dependent Normalization Framework of Visual Bottom-up Attention

  1. Shiyu Wang
  2. Ling Huang
  3. Qinglin Chen
  4. Jingyi Wang
  5. Siting Xu
  6. Xilin Zhang

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    The study claims to demonstrate an interplay between awareness and bottom-up attention and explains their joint effects within an established normalization framework. How awareness fits into current computational theory is an important and timely undertaking that has a far-reaching impact on our understanding of visual and cognitive function. Although the study uses control experiments and analyses to reinforce their claims, shortcomings in their experimental approach require further clarification and data to adequately support the study's conclusions.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Although bottom-up attention can improve visual performance with and without awareness to the exogenous cue, whether they are governed by a common neural computation remains unclear. Using a modified Posner paradigm with backward masking, we found that the cueing effect displayed a monotonic gradient profile (Gaussian-like), both with and without awareness, whose scope, however, was significantly wider with than without awareness. This awareness-dependent scope offered us a unique opportunity to change the relative size of the attention field to the stimulus, differentially modulating the gain of attentional selection, as proposed by the normalization model of attention. Therefore, for each human subject (male and female), the stimulus size was manipulated as their respective mean attention fields with and without awareness while stimulus contrast was varied in a spatial cueing task. By measuring the gain pattern of contrast-response functions on the spatial cueing effect derived by visible or invisible cues, we observed changes in the cueing effect consonant with changes in contrast gain for visible cues and response gain for invisible cues. Importantly, a complementary analysis confirmed that subjects' awareness-dependent attention fields can be simulated by using the normalization model of attention. Together, our findings indicate an awareness-dependent normalization framework of visual bottom-up attention, placing a necessary constraint, namely, awareness, on our understanding of the neural computations underlying visual attention.

SIGNIFICANCE STATEMENT Bottom-up attention is known to improve visual performance with and without awareness. We discovered that manipulating subjects' awareness can modulate their attention fields of visual bottom-up attention, which offers a unique opportunity to regulate its normalization processes. On the one hand, by measuring the gain pattern of contrast-response functions on the spatial cueing effect derived by visible or invisible cues, we observed changes in the cueing effect consonant with changes in contrast gain for visible cues and response gain for invisible cues. On the other hand, by using the normalization model of attention, subjects' awareness-dependent attention fields can be simulated successfully. Our study supports important predictions of the normalization model of visual bottom-up attention and further reveals its dependence on awareness.

Article activity feed

  1. eLife

    Evaluation Summary:

    The study claims to demonstrate an interplay between awareness and bottom-up attention and explains their joint effects within an established normalization framework. How awareness fits into current computational theory is an important and timely undertaking that has a far-reaching impact on our understanding of visual and cognitive function. Although the study uses control experiments and analyses to reinforce their claims, shortcomings in their experimental approach require further clarification and data to adequately support the study's conclusions.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 agreed to share their name with the authors.)

  2. eLife

    Reviewer #1 (Public Review):

    This work attempts to provide a novel understanding of the neural computations driving bottom-up attention. The Reynolds & Heeger normalization model of attention (NMA) can explain a variety of physiological and psychophysical findings in the attention literature but is agnostic as to the role of awareness. This paper provides findings to fill this gap with a series of novel experiments that ultimately conclude that bottom-up attention with awareness is driven by a bigger attentional scope (attention field size) whereas without awareness leads to a smaller attentional scope consonant with changes in the attention field parameter of the NMA. Ultimately, the authors suggest an awareness dependent constraint is necessary in the NMA.

    Strengths:

    1. It addresses an important gap in our understanding of how visual attention and visual awareness interact.

    2. It draws and tests predictions from a well-established computational theory that has reconciled a variety of attentional effects on physiological and psychophysical responses. By anchoring its predictions on extant theory, the study has a principled foundation from which it can decipher the computations underlying awareness.

    Weaknesses:

    1. The construct of awareness should be defined and operationalized. It is clear that the study operationalizes awareness as the visibility of a masked cue. However, this only becomes apparent during the Results section.

    2. It is stated on pg. 3 that "it's unclear whether there is a common neural computation governing bottom-up attention-triggered improvement in visual performance with and without awareness." This point should be further elaborated. Specifically, in the current literature, what is clear and what remains unclear about the computations underlying awareness and attention? The known computational underpinnings of awareness and how the current study can add to what is known should be discussed.

    3. The behavioral protocols conflate the effects of cue and mask. In both Distribution and Normalization experiments, cue location and mask contrast (high contrast in the "invisible" condition and low contrast in the "visible" condition) are concurrently manipulated. Consequently, one cannot isolate the effect of the cue from the impact of the mask on task performance.
    Although a manipulation of cue contrast was performed during the Distribution experiments, it does not completely rule out interactions between the cue and mask. With low luminance cues, cueing effect amplitudes decreased while attentional spread (i.e., scope) appeared to remain unchanged (Figure 2A vs 2D). This pattern suggests separate influences of the cue and mask, but it remains impossible to determine how the lower visibility of the cue (and thus the lower awareness thereof) affects the distribution of attentional effects and whether cue visibility interacts with or is independent from masking stimuli effects.
    A control experiment should be provided to ensure that attention is driving the reported behavioral effects.

    4. The authors suggest an awareness dependent constraint is necessary in the NMA. However, no extension of the model is provided in their modeling efforts nor proof that a constraint is necessary to capture the awareness results.

    5. Behavioral protocols for the Normalization experiment discourage any spatial distribution of attention. Unlike the Distribution experiments where cue or target position varied, the Normalization experiments used a single cue position and a single target position on the left and right of fixation. The fundamental assumption being made is that the spatial spreads of attention (with and without awareness) are the same in the Normalization and Distribution experiments. However, this assumption is not verified nor is it consistent with previous studies that manipulated the spread of the attention field. For instance, Herrmann et al., 2010 (cited in the manuscript) show that subtle adjustments of the target's position on a trial-by-trial basis lead to changes in the spread of bottom-up attention that modulate performance in a manner consistent with the predictions of NMA. Thus, locking the target into a single position would change the spread of attention relative to when the target's position varied.
    Moreover, the conclusions of the Normalization experiments become circular without an independent verification that the spread of bottom-up attention remained identical with the Distribution experiments. For example, the study cannot conclude that attention without awareness has a smaller spread because it elicited response gain modulations. Had the Normalization experiments used a similar cueing or target uncertainty protocol as the Distributions experiments, the study's conclusions would be better supported.

    6. The contrast response functions (CRFs) are coarsely sampled within the dynamic range. As a result, any conclusions about modulations of c50 (i.e., contrast gain) lack adequate support.
    When measuring contrast gain modulations, it is critical to finely sample the CRF to obtain accurate measures of c50. Previous simulations (García-Pérez & Alcalá-Quintana, 2005 Span. J. Psychol.) recommend at least five levels of contrast within the dynamic range when using a method of constant stimuli protocol, as is utilized in the current study. However, in this study a total of five levels are tested and only one exists within the dynamic range (8% contrast) whereas the next highest (15% contrast) borders the upper asymptote. Consequently, all but one c50 estimate exist between these two contrast levels (Figure 4D & 4E).
    Drawing conclusions and extracting model parameters based on the estimate of a single data point will lead to spurious and unreliable results. A hint of this weakness is apparent in Figure 5B:
    a) The NMA fit to the data shows response gain modulations that are as large as the observed contrast gain modulations.
    b) The positive correlation between the measured bottom-up spread and the NMA's attention field size is largely driven by three data points whereas the majority of the data points show a near-zero simulated attention field size.
    Contrast gain modulations constitute half of the study's conclusions regarding awareness and the NMA. Thus, a finer sampling of the CRFs is required to lend support to the study's main conclusions.

    7. Most of the discussion focuses on ruling out possible interpretations rather than providing new insights regarding why awareness may modify bottom-up attention.

  3. eLife

    Reviewer #2 (Public Review):

    This study examined how attention, drawn to the location of a sudden stimulus, changes participants' ability to discern the orientation of stimuli shown near that location immediately after, and also how this effect depends on whether the participant is aware of the initial, sudden, stimulus or not. Strengths include the fact that that the spatial profile of such attention effects, although central to many studies of visual attention, has not previously been examined, to my knowledge, in the context of the participant's awareness of the initial stimulus. Another strength is the experiment that ties observed spatial profile of attention to the influence of test stimulus contrast, as this fits the present study in the overarching framework of the existing normalization model of attention. A weakness of the study is that it is not clear that the difference in awareness between the two conditions is actually the relevant difference, given that the conditions differ in other respects as well. A related, but different, potential weakness is that the design of Experiments 1 and 2 makes it so that the initial, sudden, stimulus (the cue) is informative as to the side of the screen at which the subsequent (test) stimulus will appear, which invites a deliberate strategy on the part of the participant, to focus on that side of the screen in the 'conscious' condition but not in the 'unconscious' condition.

  4. Version published to 10.1523/jneurosci.1110-21.2021
  5. Version published to 10.1101/2021.04.18.440351 on bioRxiv