Visual homogeneity computations in the brain enable solving property-based visual tasks
Curation statements for this article:-
Curated by eLife
eLife assessment
This study uses carefully designed experiments to generate a useful behavioural and neuroimaging dataset on visual cognition. The results provide solid evidence for the involvement of higher-order visual cortex in processing visual oddballs and asymmetry. However, the evidence provided for the very strong claims of homogeneity as a novel concept in vision science, separable from existing concepts such as target saliency, is inadequate.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Most visual tasks involve looking for specific object features. But we also often perform property-based tasks where we look for specific property in an image, such as finding an odd item, deciding if two items are same, or if an object has symmetry. How do we solve such tasks? These tasks do not fit into standard models of decision making because their underlying feature space and decision process is unclear. Using well-known principles governing multiple object representations, we show that displays with repeating elements can be distinguished from heterogeneous displays using a property we define as visual homogeneity. In behavior, visual homogeneity predicted response times on visual search, same-different and symmetry tasks. Brain imaging during visual search and symmetry tasks revealed that visual homogeneity was localized to a region in the object-selective cortex. Thus, property-based visual tasks are solved in a localized region in the brain by computing visual homogeneity.
SIGNIFICANCE STATEMENT
Most visual tasks involve looking for specific features, like finding a face in a crowd. But we also often look for a particular image property – such as finding an odd item, deciding if two items are same, or judging if an object is symmetric. How does our brain solve these disparate tasks? Here, we show that these tasks can all be solved using a simple computation over object representations in higher visual cortex, which we define as visual homogeneity.
Article activity feed
-
-
-
-
Author response:
The following is the authors’ response to the original reviews.
eLife assessment:
This study uses carefully designed experiments to generate a useful behavioural and neuroimaging dataset on visual cognition. The results provide solid evidence for the involvement of higher-order visual cortex in processing visual oddballs and asymmetry. However, the evidence provided for the very strong claims of homogeneity as a novel concept in vision science, separable from existing concepts such as target saliency, is inadequate.
We appreciate the positive and balanced assessment from the reviewers. We agree that visual homogeneity is similar to existing concepts such as target saliency. We have tried our best to articulate our rationale for defining it as a novel concept. However, the debate about whether visual homogeneity is …
Author response:
The following is the authors’ response to the original reviews.
eLife assessment:
This study uses carefully designed experiments to generate a useful behavioural and neuroimaging dataset on visual cognition. The results provide solid evidence for the involvement of higher-order visual cortex in processing visual oddballs and asymmetry. However, the evidence provided for the very strong claims of homogeneity as a novel concept in vision science, separable from existing concepts such as target saliency, is inadequate.
We appreciate the positive and balanced assessment from the reviewers. We agree that visual homogeneity is similar to existing concepts such as target saliency. We have tried our best to articulate our rationale for defining it as a novel concept. However, the debate about whether visual homogeneity is novel or related to existing concepts is completely beside the point, since that is not the key contribution of our study.
Our key contribution is our quantitative model for how the brain could be solving generic visual tasks by operating on a feature space. In the literature there are no theories regarding the decision-making process by which the brain could be solving generic visual tasks. In fact, oddball search tasks, same-different tasks and symmetry tasks are never even mentioned in the same study because it is tacitly assumed that the underlying processes are completely different! Our work brings together these disparate tasks by proposing a specific computation that enables the brain to solve both types of tasks and providing evidence for it. This specific computation is a well-defined, falsifiable model that will need to be replicated, elaborated and refined by future studies.
Public Reviews:
Reviewer #1 (Public Review):
Summary:
The authors define a new metric for visual displays, derived from psychophysical response times, called visual homogeneity (VH). They attempt to show that VH is explanatory of response times across multiple visual tasks. They use fMRI to find visual cortex regions with VH-correlated activity. On this basis, they declare a new visual region in the human brain, area VH, whose purpose is to represent VH for the purpose of visual search and symmetry tasks.
Thank you for your concise summary. We appreciate your careful reading and thoughtful and constructive comments.
Strengths:
The authors present carefully designed experiments, combining multiple types of visual judgments and multiple types of visual stimuli with concurrent fMRI measurements. This is a rich dataset with many possibilities for analysis and interpretation.
Thank you for your accurate assessment of the strengths of our study.
Weaknesses:
The datasets presented here should provide a rich basis for analysis. However, in this version of the manuscript, I believe that there are major problems with the logic underlying the authors' new theory of visual homogeneity (VH), with the specific methods they used to calculate VH, and with their interpretation of psychophysical results using these methods. These problems with the coherency of VH as a theoretical construct and metric value make it hard to interpret the fMRI results based on searchlight analysis of neural activity correlated with VH.
We appreciate your concerns, and have tried our best to respond to them fully against your specific concerns below.
In addition, the large regions of VH correlations identified in Experiments 1 and 2 vs. Experiments 3 and 4 are barely overlapping. This undermines the claim that VH is a universal quantity, represented in a newly discovered area of the visual cortex, that underlies a wide variety of visual tasks and functions.
We agree with you that the VH regions defined using symmetry task and search task do not overlap completely (as we have shown in Figure S13). However this is to be expected for several reasons. First, the images in the symmetry task were presented at fixation, whereas the images in the visual search task were presented peripherally. Second, the lack of overlap could be due to variations across individuals. Indeed, considerable individual variability has been observed in the location of category-selective regions such as VWFA (Glezer and Riesenhuber 2013) and FFA (Weiner and Grill-Spector, 2012). We propose that testing the same participants on both search and symmetry tasks would reveal overlapping VH regions. We now acknowledge these issues in the Results (p. 26).
Maybe I have missed something, or there is some flaw in my logic. But, absent that, I think the authors should radically reconsider their theory, analyses, and interpretations, in light of the detailed comments below, to make the best use of their extensive and valuable datasets combining behavior and fMRI. I think doing so could lead to a much more coherent and convincing paper, albeit possibly supporting less novel conclusions.
We appreciate your concerns. We have tried our best to respond to them fully against your specific concerns below.
THEORY AND ANALYSIS OF VH
(1) VH is an unnecessary, complex proxy for response time and target-distractor similarity. VH is defined as a novel visual quality, calculable for both arrays of objects (as studied in Experiments 1-3) and individual objects (as studied in Experiment 4). It is derived from a center-to-distance calculation in a perceptual space. That space in turn is derived from the multi-dimensional scaling of response times for target-distractor pairs in an oddball detection task (Experiments 1 and 2) or in a same-different task (Experiments 3 and 4).
The above statements are not entirely correct. Experiments 1 & 3 are oddball visual search experiments. Their purpose was to estimate the underlying perceptual space of objects.
Proximity of objects in the space is inversely proportional to response times for arrays in which they were paired. These response times are higher for more similar objects. Hence, proximity is proportional to similarity. This is visible in Fig. 2B as the close clustering of complex, confusable animal shapes.
VH, i.e. distance-to-center, for target-present arrays, is calculated as shown in Fig. 1C, based on a point on the line connecting the target and distractors. The authors justify this idea with previous findings that responses to multiple stimuli are an average of responses to the constituent individual stimuli. The distance of the connecting line to the center is inversely proportional to the distance between the two stimuli in the pair, as shown in Fig. 2D. As a result, VH is inversely proportional to the distance between the stimuli and thus to stimulus similarity and response times. But this just makes VH a highly derived, unnecessarily complex proxy for target-distractor similarity and response time. The original response times on which the perceptual space is based are far more simple and direct measures of similarity for predicting response times.
We agree that VH brings no explanatory power to target-present searches, since target-present response times are a direct estimate of target-distractor similarity. However, we are additionally explaining target-absent response times. Target-absent response times are well known to vary systematically with image properties, but why they do so have not been clear in the literature.
Our key conceptual advance lies in relating the neural response to a search array to the neural response of the constituent elements, and in proposing a decision variable using which participants can make both target-present and target-absent judgements on any search array.
(2) The use of VH derived from Experiment 1 to predict response times in Experiment 2 is circular and does not validate the VH theory.
The use of VH, a response time proxy, to predict response times in other, similar tasks, using the same stimuli, is circular. In effect, response times are being used to predict response times across two similar experiments using the same stimuli. Experiment 1 and the target present condition of Experiment 2 involve the same essential task of oddball detection. The results of Experiment 1 are converted into VH values as described above, and these are used to predict response times in Experiment 2 (Fig. 2F). Since VH is a derived proxy for response values in Experiment 1, this prediction is circular, and the observed correlation shows only consistency between two oddball detection tasks in two experiments using the same stimuli.
We agree that it would be circular to use oddball search times in Experiment 1 to explain only target-present search times in Experiment 2, since they basically involve the same searches. However, we are explaining both target-present and target-absent search times in a unified framework; systematic variations in target-absent search times have been noted in the literature but never really explained. One could still simply say that target-absent search times are some function of the target-present search times, but this still doesn’t provide an explanation for how participants are making target-present and absent decisions. The existing literature contains models for how visual search might occur for a specific target and distractor but does not elucidate how participants might perform generic visual search where target and distractors are not known in advance.
Our key conceptual advance lies in relating the neural response to a search array to the neural response of the constituent elements, and in proposing a decision variable using which participants can make both target-present and target-absent judgements on any search array.
(3) The negative correlation of target-absent response times with VH as it is defined for target-absent arrays, based on the distance of a single stimulus from the center, is uninterpretable without understanding the effects of center-fitting. Most likely, center-fitting and the different VH metrics for target-absent trials produce an inverse correlation of VH with target-distractor similarity.
We see no cause for concern with the center-fitting procedure, for several reasons. First, the best-fitting center remained stable despite many randomly initialized starting points. Second, the best-fitting center derived from one set of objects was able to predict the target-absent and target-present responses of another set of objects. Finally, the VH obtained for each object (i.e. distance from the best-fitting center) is strongly correlated with the average distance of that object from all other objects (Figure S1A). We have now clarified this in the Results (p. 11).
The construction of the VH perceptual space also involves fitting a "center" point such that distances to center predict response times as closely as possible. The effect of this fitting process on distance-to-center values for individual objects or clusters of objects is unknowable from what is presented here. These effects would depend on the residual errors after fitting response times with the connecting line distances. The center point location and its effects on the distance-to-center of single objects and object clusters are not discussed or reported here.
While it is true that the optimal center needs to be found by fitting to the data, there no particular mystery to the algorithm: we are simply performing a standard gradient-descent to maximize the fit to the data. We have described the algorithm clearly and are making our codes public. We find the algorithm to yield stable optimal centers despite many randomly initialized starting points. We find the optimal center to be able to predict responses to entirely novel images that were excluded during model training. We are making no assumption about the location of centre with respect to individual points. Therefore, we see no cause for concern regarding the center-finding algorithm.
Yet, this uninterpretable distance-to-center of single objects is chosen as the metric for VH of target-absent displays (VHabsent). This is justified by the idea that arrays of a single stimulus will produce an average response equal to one stimulus of the same kind. However, it is not logically clear why response strength to a stimulus should be a metric for homogeneity of arrays constructed from that stimulus, or even what homogeneity could mean for a single stimulus from this set. It is not clear how this VHabsent metric based on single stimuli can be equated to the connecting line VH metric for stimulus pairs, i.e. VHpresent, or how both could be plotted on a single continuum.
Most visual tasks, such as finding an animal, are thought to involve building a decision boundary on some underlying neural representation. Even visual search has been portrayed as a signal-detection problem where a particular target is to be discriminated from a distractor. However none of these formulations work in the case of generic visual tasks, where the target and distractor identities are unknown. We are proposing that, when we view a search array, the neural response to the search array can be deduced from the neural responses to the individual elements using well known rules, and that decisions about an oddball target being present or absent can be made by computing the distance of this neural response from some canonical mean firing rate of a population of neurons. This distance to center computation is what we denote as visual homogeneity. We have revised our manuscript throughout to make this clearer and we hope that this helps you understand the logic better.
It is clear, however, what should be correlated with difficulty and response time in the target-absent trials, and that is the complexity of the stimuli and the numerosity of similar distractors in the overall stimulus set. The complexity of the target, similarity with potential distractors, and the number of such similar distractors all make ruling out distractor presence more difficult. The correlation seen in Fig. 2G must reflect these kinds of effects, with higher response times for complex animal shapes with lots of similar distractors and lower response times for simpler round shapes with fewer similar distractors.
You are absolutely correct that the stimulus complexity should matter, but there are no good measures for stimulus complexity. But considering what factors are correlated with target-absent response times is entirely different from asking what decision variable or template is being used by participants to solve the task.
The example points in Fig. 2G seem to bear this out, with higher response times for the deer stimulus (complex, many close distractors in the Fig. 2B perceptual space) and lower response times for the coffee cup (simple, few close distractors in the perceptual space). While the meaning of the VH scale in Fig. 2G, and its relationship to the scale in Fig. 2F, are unknown, it seems like the Fig. 2G scale has an inverse relationship to stimulus complexity, in contrast to the expected positive relationship for Fig. 2F. This is presumably what creates the observed negative correlation in Fig. 2G.
Taken together, points 1-3 suggest that VHpresent and VHabsent are complex, unnecessary, and disconnected metrics for understanding target detection response times. The standard, simple explanation should stand. Task difficulty and response time in target detection tasks, in both present and absent trials, are positively correlated with target-distractor similarity.
Respectfully, we disagree with your assessment. Your last point is not logically consistent though: response times for target-absent trials cannot be correlated with any target-distractor similarity since there is no target in the first place in a target-absent array. We have shown that target-absent response times are in fact, independent of experimental context, which means that they index an image property that is independent of any reference target (Results, p. 15; Section S4). This property is what we define as visual homogeneity.
I think my interpretations apply to Experiments 3 and 4 as well, although I find the analysis in Fig. 4 especially hard to understand. The VH space in this case is based on Experiment 3 oddball detection in a stimulus set that included both symmetric and asymmetric objects. However, the response times for a very different task in Experiment 4, a symmetric/asymmetric judgment, are plotted against the axes derived from Experiment 3 (Fig. 4F and 4G). It is not clear to me why a measure based on oddball detection that requires no use of symmetry information should be predictive of within-stimulus symmetry detection response times. If it is, that requires a theoretical explanation not provided here.
We are using an oddball detection task to estimate perceptual dissimilarity between objects, and construct the underlying perceptual representation of both symmetric and asymmetric objects. This enabled us to then ask if some distance-to-center computation can explain response times in a symmetry detection task, and obtain an answer in the affirmative. We have reworked the text to make this clear.
(4) Contrary to the VH theory, same/different tasks are unlikely to depend on a decision boundary in the middle of a similarity or homogeneity continuum.
We have provided empirical proof for our claims, by showing that target-present response times in a visual search task are correlated with “different” responses in the same-different task, and that target-absent response times in the visual search task are correlated with “same” responses in the same-different task (Section S3).
The authors interpret the inverse relationship of response times with VHpresent and VHabsent, described above, as evidence for their theory. They hypothesize, in Fig. 1G, that VHpresent and VHabsent occupy a single scale, with maximum VHpresent falling at the same point as minimum VHabsent. This is not borne out by their analysis, since the VHpresent and VHabsent value scales are mainly overlapping, not only in Experiments 1 and 2 but also in Experiments 3 and 4. The authors dismiss this problem by saying that their analyses are a first pass that will require future refinement. Instead, the failure to conform to this basic part of the theory should be a red flag calling for revision of the theory.
We respectfully disagree – by no means did we dismiss this problem! In fact, we have explicitly acknowledged this by saying that VH does not explain all the variance in the response times, but nonetheless explains substantial variance and might form the basis for an initial guess or a fast response. The remaining variance might be explained by processes that involve more direct scrutiny. Please see Results, page 10 & 22.
The reason for this single scale is that the authors think of target detection as a boundary decision task, along a single scale, with a decision boundary somewhere in the middle, separating present and absent. This model makes sense for decision dimensions or spaces where there are two categories (right/left motion; cats vs. dogs), separated by an inherent boundary (equal left/right motion; training-defined cat/dog boundary). In these cases, there is less information near the boundary, leading to reduced speed/accuracy and producing a pattern like that shown in Fig. 1G.
The key conceptual advance of our study is that we show that even target/present, same/different or symmetry judgements can be fit into the standard decision-making framework.
This logic does not hold for target detection tasks. There is no inherent middle point boundary between target present and target absent. Instead, in both types of trials, maximum information is present when the target and distractors are most dissimilar, and minimum information is present when the target and distractors are most similar. The point of greatest similarity occurs at the limit of any metric for similarity. Correspondingly, there is no middle point dip in information that would produce greater difficulty and higher response times. Instead, task difficulty and response times increase monotonically with the similarity between targets and distractors, for both target present and target absent decisions. Thus, in Figs. 2F and 2G, response times appear to be highest for animals, which share the largest numbers of closely similar distractors.
Unfortunately, your logic does not boil down to any quantitative account, since you are using vague terms like “maximum information”. Further, any argument based solely on item similarity to explain visual search or symmetry responses cannot explain systematic variations observed for target-absent arrays and for symmetric objects, for the reasons below.
If target-distractor dissimilarity were the sole driver of response times, target-absent judgements should always take the longest time since the target and distractor have zero similarity, with no variation from one image to another. This account does not explain why target-absent response times vary so systematically.
Similarly, if symmetry judgements are solely based on comparing the dissimilarity between two halves of an object, there should be no variation in the response times of symmetric objects since the dissimilarity between their two halves is zero. However we do see systematic variation in the response times to symmetric objects.
DEFINITION OF AREA VH USING fMRI
(1) The area VH boundaries from different experiments are nearly completely non-overlapping.
In line with their theory that VH is a single continuum with a decision boundary somewhere in the middle, the authors use fMRI searchlight to find an area whose responses positively correlate with homogeneity, as calculated across all of their target present and target absent arrays. They report VH-correlated activity in regions anterior to LO. However, the VH defined by symmetry Experiments 3 and 4 (VHsymmetry) is substantially anterior to LO, while the VH defined by target detection Experiments 1 and 2 (VHdetection) is almost immediately adjacent to LO. Fig. S13 shows that VHsymmetry and VHdetection are nearly non-overlapping. This is a fundamental problem with the claim of discovering a new area that represents a new quantity that explains response times across multiple visual tasks. In addition, it is hard to understand why VHsymmetry does not show up in a straightforward subtraction between symmetric and asymmetric objects, which should show a clear difference in homogeneity. • Actually VHsymmetry is apparent even in a simple subtraction between symmetric and asymmetric objects (Figure S10). The VH regions identified using the visual search task and symmetry task have a partial overlap, not zero overlap as you are incorrectly claiming.
We have noted that it is not straightforward to interpret the overlap, since there are many confounding factors. One reason could simply be that the stimuli in the symmetry task were presented at fixation, whereas the visual search arrays contained items exclusively in the periphery. Another that the participants in the two tasks were completely different, and the lack of overlap is simply due to inter-individual variability. Testing the same participants in two tasks using similar stimuli would be ideal but this is outside the scope of this study. We have acknowledged these issues in the Results (p. 26) and in the Supplementary Material (Section S8).
(2) It is hard to understand how neural responses can be correlated with both VHpresent and VHabsent.
The main paper results for VHdetection are based on both target-present and target-absent trials, considered together. It is hard to interpret the observed correlations, since the VHpresent and VHabsent metrics are calculated in such different ways and have opposite correlations with target similarity, task difficulty, and response times (see above). It may be that one or the other dominates the observed correlations. It would be clarifying to analyze correlations for target-present and target-absent trials separately, to see if they are both positive and correlated with each other.
Thanks. The positive correlation between VH and neural response holds even when we do the analysis separately for target-present and -absent searches (correlation between neural response in VH region and visual homogeneity (n = 32, r = 0.66, p < 0.0005 for target-present searches & n = 32, r = 0.56, p < 0.005 for target-absent searches).
(3) The definition of the boundaries and purpose of a new visual area in the brain requires circumspection, abundant and convergent evidence, and careful controls.
Even if the VH metric, as defined and calculated by the authors here, is a meaningful quantity, it is a bold claim that a large cortical area just anterior to LO is devoted to calculating this metric as its major task. Vision involves much more than target detection and symmetry detection. The cortex anterior to LO is bound to perform a much wider range of visual functionalities. If the reported correlations can be clarified and supported, it would be more circumspect to treat them as one byproduct of unknown visual processing in the cortex anterior to LO, rather than treating them as the defining purpose for a large area of the visual cortex.
We totally agree with you that reporting a new brain region would require careful interpretation and abundant and converging evidence. However, this requires many studies worth of work, and historically category-selective regions like the FFA have achieved consensus only after they were replicated and confirmed across many studies. We believe our proposal for the computation of a quantity like visual homogeneity is conceptually novel, and our study represents a first step that provides some converging evidence (through replicable results across different experiments) for such a region. We have reworked our manuscript to make this point clearer (Discussion, p 32).
Reviewer #2 (Public Review):
Summary:
This study proposes visual homogeneity as a novel visual property that enables observers perform to several seemingly disparate visual tasks, such as finding an odd item, deciding if two items are the same, or judging if an object is symmetric. In Experiment 1, the reaction times on several objects were measured in human subjects. In Experiment 2, the visual homogeneity of each object was calculated based on the reaction time data. The visual homogeneity scores predicted reaction times. This value was also correlated with the BOLD signals in a specific region anterior to LO. Similar methods were used to analyze reaction time and fMRI data in a symmetry detection task. It is concluded that visual homogeneity is an important feature that enables observers to solve these two tasks.
Strengths:
(1) The writing is very clear. The presentation of the study is informative.
(2) This study includes several behavioral and fMRI experiments. I appreciate the scientific rigor of the authors.
We are grateful to you for your balanced assessment and constructive comments.
Weaknesses:
(1) My main concern with this paper is the way visual homogeneity is computed. On page 10, lines 188-192, it says: "we then asked if there is any point in this multidimensional representation such that distances from this point to the target-present and target-absent response vectors can accurately predict the target-present and target-absent response times with a positive and negative correlation respectively (see Methods)". This is also true for the symmetry detection task. If I understand correctly, the reference point in this perceptual space was found by deliberating satisfying the negative and positive correlations in response times. And then on page 10, lines 200-205, it shows that the positive and negative correlations actually exist. This logic is confusing. The positive and negative correlations emerge only because this method is optimized to do so. It seems more reasonable to identify the reference point of this perceptual space independently, without using the reaction time data. Otherwise, the inference process sounds circular. A simple way is to just use the mean point of all objects in Exp 1, without any optimization towards reaction time data.
We disagree with you since the same logic applies to any curve-fitting procedure. When we fit data to a straight line, we are finding the slope and intercept that minimizes the error between the data and the straight line, but we would hardly consider the process circular when a good fit is achieved – in fact we take it as a confirmation that the data can be fit linearly. In the same vein, we would not have observed a good fit to the data, if there did not exist any good reference point relative to which the distances of the target-present and target-absent search arrays predicted these response times.
In Section S1, we have already reported that the visual homogeneity estimates for each object is strongly correlated with the average distance of each object to all other objects (r = 0.84, p<0.0005, Figure S1). Second, to confirm that the results we obtained are not due to overfitting, we have already reported a cross-validation analysis, where we removed all searches involving a particular image and predicted these response times using visual homogeneity. This too revealed a significant model correlation confirming that our results are not due to overfitting.
(2) On page 11, lines 214-221. It says: "these findings are non-trivial for several reasons". However, the first reason is confusing. It is unclear to me why "it suggests that there are highly specific computations that can be performed on perceptual space to solve oddball tasks". In fact, these two sentences provide no specific explanation for the results.
We have now revised the text to make it clearer (Results, p. 11).
(3) The second reason is interesting. Reaction times in target-present trials can be easily explained by target-distractor similarity. But why does reaction time vary substantially across target-absent stimuli? One possible explanation is that the objects that are distant from the feature distribution elicit shorter reaction times. Here, all objects constitute a statistical distribution in the feature (perceptual) space. There is certainly a mean of this distribution. Some objects look like outliers and these outliers elicit shorter reaction times in the target-absent trials because outlier detection is very salient.
One might argue that the above account is merely a rephrasing of the idea of visual homogeneity proposed in this study. If so, feature saliency is not a new account. In other words, the idea of visual homogeneity is another way of reiterating the old feature saliency theory.
Thank you for this interesting point. We don’t necessarily see a contradiction. However, we are proposing a quantitative decision variable that the brain could be using to make target present/absent judgements.
(4) One way to reject the feature saliency theory is to compare the reaction times of the objects that are very different from other objects (i.e., no surrounding objects in the perceptual space, e.g., the wheel in the lower right corner of Fig. 2B) with the objects that are surrounded by several similar objects (e.g., the horse in the upper part of Fig. 2B). Also, please choose the two objects with similar distance from the reference point. I predict that the latter will elicit longer reaction times because they can be easily confounded by surrounding similar objects (i.e., four-legged horses can be easily confounded by four-legged dogs). If the density of object distribution per se influences the visual homogeneity score, I would say that the "visual homogeneity" is essentially another way of describing the distributional density of the perceptual space.
We agree with you, and we have indeed found that visual homogeneity estimates from our model are highly correlated with the average distance of an object relative to all other objects. However, we performed several additional experiments to elucidate the nature of target-absent response times. We find that they are unaffected by whether these searches are performed in the midst of similar or dissimilar objects (Section S4, Experiment S6), and even when the same searches are performed among nearby sets of objects with completely uncorrelated average distances (Section S4, Experiment S7). We have now reworked the text to make this clearer.
(5) The searchlight analysis looks strange to me. One can easily perform a parametric modulation by setting visual homogeneity as the trial-by-trial parametric modulator and reaction times as a covariate. This parametric modulation produces a brain map with the correlation of every voxel in the brain. On page 17 lines 340-343, it is unclear to me what the "mean activation" is.
We have done something similar. For each region we took the mean activation at each voxel as the average activation 3x3x3 voxel neighborhood in the brain, and took its correlation with visual homogeneity. We have now reworked this to make it clearer (Results, p. 16).
Minor points
(1) In the intro, it says: "using simple neural rules..." actually it is very confusing what "neural rules" are here. Better to change it to "computational principles" or "neural network models"??
We have now replaced this with “using well-known principles governing multiple object representations”.
(2) In the intro, it says: "while machine vision algorithms are extremely successful in solving feature-based tasks like object categorization (Serre, 2019), they struggle to solve these generic tasks (Kim et al., 2018; Ricci et al. 2021). These are not generic tasks. They are just a specific type of visual task-judging relationship between multiple objects. Moreover, a large number of studies in machine vision have shown that DNNs are capable of solving these tasks and even more difficult tasks. Two survey papers are listed here.
Wu, Q., Teney, D., Wang, P., Shen, C., Dick, A., & Van Den Hengel, A. (2017). Visual question answering: A survey of methods and datasets. Computer Vision and Image Understanding, 163, 21-40.
Małkiński, M., & Mańdziuk, J. (2022). Deep Learning Methods for Abstract Visual Reasoning: A Survey on Raven's Progressive Matrices. arXiv preprint arXiv:2201.12382.
Thank you for sharing these references. In fact, a recent study has shown that specific deep networks can indeed solve the same-different task (Tartaglini et al, 2023). However our broader point remains that the same-different or other such visual tasks are non-trivial for machine vision algorithms.
Reviewer #1 (Recommendations For The Authors):
Nothing to add to the public review. If my concerns turn out to be invalid, I apologize and will happily accept correction. If they are valid, I hope they will point toward a new version of this paper that optimizes the insights to be gained from this impressive dataset.
Reviewer #2 (Recommendations For The Authors):
My suggestions are as follows:
(1) Analyze the fMRI data using the parametric modulation approach first at the single-subject level and then perform group analysis.
To clarify, we have obtained image-level activations from each subject, and used it for all our analyses.
(2) Think about a way to redefine visual homogeneity from a purely image-computable approach. In other words, visual homogeneity should be first defined as an image feature that is independent of any empirical response data. And then use the visual homogeneity scores to predict reaction times.
While we understand what you mean, any image-computable representation such as from a deep network may carry its own biases and may not be an accurate representation of the underlying object representation. By contrast, neural dissimilarities in the visual cortex are strongly predictive of visual search oddball response times. That is why we used visual search oddball response times as a proxy for the underlying neural representation, and then asked whether some decision variable can be derived from this representation to explain both target present and absent judgements in visual search.
-
eLife assessment
This study uses carefully designed experiments to generate a useful behavioural and neuroimaging dataset on visual cognition. The results provide solid evidence for the involvement of higher-order visual cortex in processing visual oddballs and asymmetry. However, the evidence provided for the very strong claims of homogeneity as a novel concept in vision science, separable from existing concepts such as target saliency, is inadequate.
-
Reviewer #1 (Public Review):
Summary:
The authors define a new metric for visual displays, derived from psychophysical response times, called visual homogeneity (VH). They attempt to show that VH is explanatory of response times across multiple visual tasks. They use fMRI to find visual cortex regions with VH-correlated activity. On this basis, they declare a new visual region in human brain, area VH, whose purpose is to represent VH for the purpose of visual search and symmetry tasks.
Strengths:
The authors present carefully designed experiments, combining multiple types of visual judgments and multiple types of visual stimuli with concurrent fMRI measurements. This is a rich dataset with many possibilities for analysis and interpretation.
Weaknesses:
The datasets presented here should provide a rich basis for analysis. However, in …
Reviewer #1 (Public Review):
Summary:
The authors define a new metric for visual displays, derived from psychophysical response times, called visual homogeneity (VH). They attempt to show that VH is explanatory of response times across multiple visual tasks. They use fMRI to find visual cortex regions with VH-correlated activity. On this basis, they declare a new visual region in human brain, area VH, whose purpose is to represent VH for the purpose of visual search and symmetry tasks.
Strengths:
The authors present carefully designed experiments, combining multiple types of visual judgments and multiple types of visual stimuli with concurrent fMRI measurements. This is a rich dataset with many possibilities for analysis and interpretation.
Weaknesses:
The datasets presented here should provide a rich basis for analysis. However, in this version of the manuscript, I believe that there are major problems with the logic underlying the authors' new theory of visual homogeneity (VH), with the specific methods they used to calculate VH, and with their interpretation of psychophysical results using these methods. These problems with the coherency of VH as a theoretical construct and metric value make it hard to interpret the fMRI results based on searchlight analysis of neural activity correlated with VH. In addition, the large regions of VH correlations identified in Experiments 1 and 2 vs. Experiments 3 and 4 are barely overlapping. This undermines the claim that VH is a universal quantity, represented in a newly discovered area of visual cortex, that underlies a wide variety of visual tasks and functions.
Maybe I have missed something, or there is some flaw in my logic. But, absent that, I think the authors should radically reconsider their theory, analyses, and interpretations, in light of detailed comments below, in order to make the best use of their extensive and valuable datasets combining behavior and fMRI. I think doing so could lead to a much more coherent and convincing paper, albeit possibly supporting less novel conclusions.
THEORY AND ANALYSIS OF VH
(1) VH is an unnecessary, complex proxy for response time and target-distractor similarity.
VH is defined as a novel visual quality, calculable for both arrays of objects (as studied in Experiments 1-3) and individual objects (as studied in Experiment 4). It is derived from a center-to-distance calculation in a perceptual space. That space in turn is derived from multi-dimensional scaling of response times for target-distractor pairs in an oddball detection task (Experiments 1 and 2) or in a same different task (Experiments 3 and 4). Proximity of objects in the space is inversely proportional to response times for arrays in which they were paired. These response times are higher for more similar objects. Hence, proximity is proportional to similarity. This is visible in Fig. 2B as the close clustering of complex, confusable animal shapes.
VH, i.e. distance-to-center, for target-present arrays is calculated as shown in Fig. 1C, based on a point on the line connecting target and distractors. The authors justify this idea with previous findings that responses to multiple stimuli are an average of responses to the constituent individual stimuli. The distance of the connecting line to the center is inversely proportional to the distance between the two stimuli in the pair, as shown in Fig. 2D. As a result, VH is inversely proportional to distance between the stimuli and thus to stimulus similarity and response times. But this just makes VH a highly derived, unnecessarily complex proxy for target-distractor similarity and response time. The original response times on which the perceptual space is based are far more simple and direct measures of similarity for predicting response times.
(2) The use of VH derived from Experiment 1 to predict response times in Experiment 2 is circular and does not validate the VH theory.
The use of VH, a response time proxy, to predict response times in other, similar tasks, using the same stimuli, is circular. In effect, response times are being used to predict response times across two similar experiments using the same stimuli. Experiment 1 and the target present condition of Experiment 2 involve the same essential task of oddball detection. The results of Experiment 1 are converted into VH values as described above, and these are used to predict response times in experiment 2 (Fig. 2F). Since VH is a derived proxy for response values in Experiment 1, this prediction is circular, and the observed correlation shows only consistency between two oddball detection tasks in two experiments using the same stimuli.
(3) The negative correlation of target-absent response times with VH as it is defined for target-absent arrays, based on distance of a single stimulus from center, is uninterpretable without understanding the effects of center-fitting. Most likely, center-fitting and the different VH metric for target-absent trials produce an inverse correlation of VH with target-distractor similarity.
The construction of the VH perceptual space also involves fitting a "center" point such that distances to center predict response times as closely as possible. The effect of this fitting process on distance-to-center values for individual objects or clusters of objects is unknowable from what is presented here. These effects would depend on the residual errors after fitting response times with the connecting line distances. The center point location and its effects on distance-to-center of single objects and object clusters are not discussed or reported here.
Yet, this uninterpretable distance-to-center of single objects is chosen as the metric for VH of target-absent displays (VHabsent). This is justified by the idea that arrays of a single stimulus will produce an average response equal to one stimulus of the same kind. But it is not logically clear why response strength to a stimulus should be a metric for homogeneity of arrays constructed from that stimulus, or even what homogeneity could mean for a single stimulus from this set. And it is not clear how this VHabsent metric based on single stimuli can be equated to the connecting line VH metric for stimulus pairs, i.e. VHpresent, or how both could be plotted on a single continuum.
It is clear, however, what *should* be correlated with difficulty and response time in the target-absent trials, and that is the complexity of the stimuli and the numerosity of similar distractors in the overall stimulus set. Complexity of the target, similarity with potential distractors, and number of such similar distractors all make ruling out distractor presence more difficult. The correlation seen in Fig. 2G must reflect these kinds of effects, with higher response times for complex animal shapes with lots of similar distractors and lower response times for simpler round shapes with fewer similar distractors.
The example points in Fig. 2G seem to bear this out, with higher response times for the deer stimulus (complex, many close distractors in the Fig. 2B perceptual space) and lower response times for the coffee cup (simple, few close distractors in the perceptual space). While the meaning of the VH scale in Fig. 2G, and its relationship to the scale in Fig. 2F, are unknown, it seems like the Fig. 2G scale has an inverse relationship to stimulus complexity, in contrast to the expected positive relationship for Fig. 2F. This is presumably what creates the observed negative correlation in Fig. 2G.
Taken together, points 1-3 suggest that VHpresent and VHabsent are complex, unnecessary, and disconnected metrics for understanding target detection response times. The standard, simple explanation should stand. Task difficulty and response time in target detection tasks, in both present and absent trials, are positively correlated with target-distractor similarity.
I think my interpretations apply to Experiments 3 and 4 as well, although I find the analysis in Fig. 4 especially hard to understand. The VH space in this case is based on Experiment 3 oddball detection in a stimulus set that included both symmetric and asymmetric objects. But the response times for a very different task in Experiment 4, a symmetric/asymmetric judgment, are plotted against the axes derived from Experiment 3 (Fig. 4F and 4G). It is not clear to me why a measure based on oddball detection that requires no use of symmetry information should be predictive of within-stimulus symmetry detection response times. If it is, that requires a theoretical explanation not provided here.
(4) Contrary to the VH theory, same/different tasks are unlikely to depend on a decision boundary in the middle of a similarity or homogeneity continuum.
The authors interpret the inverse relationship of response times with VHpresent and VHabsent, described above, as evidence for their theory. They hypothesize, in Fig. 1G, that VHpresent and VHabsent occupy a single scale, with maximum VHpresent falling at the same point as minimum VHabsent. This is not borne out by their analysis, since the VHpresent and VHabsent value scales are mainly overlapping, not only in Experiments 1 and 2 but also in Experiments 3 and 4. The authors dismiss this problem by saying that their analyses are a first pass that will require future refinement. Instead, the failure to conform to this basic part of the theory should be a red flag calling for revision of the theory.
The reason for this single scale is that the authors think of target detection as a boundary decision task, along a single scale, with a decision boundary somewhere in the middle, separating present and absent. This model makes sense for decision dimensions or spaces where there are two categories (right/left motion; cats vs. dogs), separated by an inherent boundary (equal left/right motion; training-defined cat/dog boundary). In these cases, there is less information near the boundary, leading to reduced speed/accuracy and producing a pattern like that shown in Fig. 1G.
This logic does not hold for target detection tasks. There is no inherent middle point boundary between target present and target absent. Instead, in both types of trial, maximum information is present when target and distractors are most dissimilar, and minimum information is present when target and distractors are most similar. The point of greatest similarity occurs at then limit of any metric for similarity. Correspondingly, there is no middle point dip in information that would produce greater difficulty and higher response times. Instead, task difficulty and response times increase monotonically with similarity between targets and distractors, for both target present and target absent decisions. Thus, in Figs. 2F and 2G, response times appear to be highest for animals, which share the largest numbers of closely similar distractors.
DEFINITION OF AREA VH USING fMRI
(1) The area VH boundaries from different experiments are nearly completely non-overlapping.
In line with their theory that VH is a single continuum with a decision boundary somewhere in the middle, the authors use fMRI searchlight to find an area whose responses positively correlate with homogeneity, as calculated across all of their target present and target absent arrays. They report VH-correlated activity in regions anterior to LO. However, the VH defined by symmetry Experiments 3 and 4 (VHsymmetry) is substantially anterior to LO, while the VH defined by target detection Experiments 1 and 2 (VHdetection) is almost immediately adjacent to LO. Fig. S13 shows that VHsymmetry and VHdetection are nearly non-overlapping. This is a fundamental problem with the claim of discovering a new area that represents a new quantity that explains response times across multiple visual tasks. In addition, it is hard to understand why VHsymmetry does not show up in a straightforward subtraction between symmetric and asymmetric objects, which should show a clear difference in homogeneity.
(2) It is hard to understand how neural responses can be correlated with both VHpresent and VHabsent.
The main paper results for VHdetection are based on both target-present and target-absent trials, considered together. It is hard to interpret the observed correlations, since the VHpresent and VHabsent metrics are calculated in such different ways and have opposite correlations with target similarity, task difficulty, and response times (see above). It may be that one or the other dominates the observed correlations. It would be clarifying to analyze correlations for target-present and target-absent trials separately, to see if they are both positive and correlated with each other.
(3) Definition of the boundaries and purpose of a new visual area in the brain requires circumspection, abundant and convergent evidence, and careful controls.
Even if the VH metric, as defined and calculated by the authors here, is a meaningful quantity, it is a bold claim that a large cortical area just anterior to LO is devoted to calculating this metric as its major task. Vision involves much more than target detection and symmetry detection. Cortex anterior to LO is bound to perform a much wider range of visual functionalities. If the reported correlations can be clarified and supported, it would be more circumspect to treat them as one byproduct of unknown visual processing in cortex anterior to LO, rather than treating them as the defining purpose for a large area of visual cortex.
-
Reviewer #3 (Public Review):
Summary:
This study proposes visual homogeneity as a novel visual property that enables observers perform to several seemingly disparate visual tasks, such as finding an odd item, deciding if two items are same, or judging if an object is symmetric. In Exp 1, the reaction times on several objects were measured in human subjects. In Exp 2, visual homogeneity of each object was calculated based on the reaction time data. The visual homogeneity scores predicted reaction times. This value was also correlated with the BOLD signals in a specific region anterior to LO. Similar methods were used to analyze reaction time and fMRI data in a symmetry detection task. It is concluded that visual homogeneity is an important feature that enables observers to solve these two tasks.
Strengths:
(1) The writing is very clear. …
Reviewer #3 (Public Review):
Summary:
This study proposes visual homogeneity as a novel visual property that enables observers perform to several seemingly disparate visual tasks, such as finding an odd item, deciding if two items are same, or judging if an object is symmetric. In Exp 1, the reaction times on several objects were measured in human subjects. In Exp 2, visual homogeneity of each object was calculated based on the reaction time data. The visual homogeneity scores predicted reaction times. This value was also correlated with the BOLD signals in a specific region anterior to LO. Similar methods were used to analyze reaction time and fMRI data in a symmetry detection task. It is concluded that visual homogeneity is an important feature that enables observers to solve these two tasks.
Strengths:
(1) The writing is very clear. The presentation of the study is informative.
(2) This study includes several behavioral and fMRI experiments. I appreciate the scientific rigor of the authors.Weaknesses:
(1) My main concern with this paper is the way visual homogeneity is computed. On page 10, lines 188-192, it says: "we then asked if there is any point in this multidimensional representation such that distances from this point to the target-present and target-absent response vectors can accurately predict the target-present and target-absent response times with a positive and negative correlation respectively (see Methods)". This is also true for the symmetry detection task. If I understand correctly, the reference point in this perceptual space was found by deliberating satisfying the negative and positive correlations in response times. And then on page 10, lines 200-205, it shows that the positive and negative correlations actually exist. This logic is confusing. The positive and negative correlations emerge only because this method is optimized to do so. It seems more reasonable to identify the reference point of this perceptual space independently, without using the reaction time data. Otherwise, the inference process sounds circular. A simple way is to just use the mean point of all objects in Exp 1, without any optimization towards reaction time data.
(2) Visual homogeneity (at least given the current from) is an unnecessary term. It is similar to distractor heterogeneity/distractor variability/distractor statics in literature. However, the authors attempt to claim it as a novel concept. The title is "visual homogeneity computations in the brain enable solving generic visual tasks". The last sentence of the abstract is "a NOVEL IMAGE PROPERTY, visual homogeneity, is encoded in a localized brain region, to solve generic visual tasks". In the significance, it is mentioned that "we show that these tasks can be solved using a simple property WE DEFINE as visual homogeneity". If the authors agree that visual homogeneity is not new, I suggest a complete rewrite of the title, abstract, significance, and introduction.
(3) Also, "solving generic tasks" is another overstatement. The oddball search tasks, same-different tasks, and symmetric tasks are only a small subset of many visual tasks. Can this "quantitative model" solve motion direction judgment tasks, visual working memory tasks? Perhaps so, but at least this manuscript provides no such evidence. On line 291, it says "we have proposed that visual homogeneity can be used to solve any task that requires discriminating between homogeneous and heterogeneous displays". I think this is a good statement. A title that says "XXXX enable solving discrimination tasks with multi-component displays" is more acceptable. The phrase "generic tasks" is certainly an exaggeration.
(4) If I understand it correctly, one of the key findings of this paper is "the response times for target-present searches were positively correlated with visual homogeneity. By contrast, the response times for target-absent searches were negatively correlated with visual homogeneity" (lines 204-207). I think the authors have already acknowledged that the positive correlation is not surprising at all because it reflects the classic target-distractor similarity effect. But the authors claim that the negative correlations in target-absent searches is the true novel finding.
(5) I would like to make it clear that this negative correlation is not new either. The seminal paper by Duncan and Humphreys (1989) has clearly stated that "difficulty increases with increased similarity of targets to nontargets and decreased similarity between nontargets" (the sentence in their abstract). Here, "similarity between nontargets" is the same as the visual homogeneity defined here. Similar effects have been shown in Duncan (1989) and Nagy, Neriani, and Young (2005). See also the inconsistent results in Nagy& Thomas, 2003, Vicent, Baddeley, Troscianko&Gilchrist, 2009.
More recently, Wei Ji Ma has systematically investigated the effects of heterogeneous distractors in visual search. I think the introduction part of Wei Ji Ma's paper (2020) provides a nice summary of this line of research.I am surprised that these references are not mentioned at all in this manuscript (except Duncan and Humphreys, 1989).
(6) If the key contribution is the quantitative model, the study should be organized in a different way. Although the findings of positive and negative correlations are not novel, it is still good to propose new models to explain classic phenomena. I would like to mention the three studies by Wei Ji Ma (see below). In these studies, Bayesian observer models were established to account for trial-by-trial behavioral responses. These computational models can also account for the set-size effect, behavior in both localization and detection tasks. I see much more scientific rigor in their studies. Going back to the quantitative model in this paper, I am wondering whether the model can provide any qualitative prediction beyond the positive and negative correlations? Can the model make qualitative predictions that differ from those of Wei Ji's model? If not, can the authors show that the model can quantitatively better account for the data than existing Bayesian models? We should evaluate a model either qualitatively or quantitatively.
(7) In my opinion, one of the advantages of this study is the fMRI dataset, which is valuable because previous studies did not collect fMRI data. The key contribution may be the novel brain region associated with display heterogeneity. If this is the case, I would suggest using a more parametric way to measure this region. For example, one can use Gabor stimuli and systematically manipulate the variations of multiple Gabor stimuli, the same logic also applies to motion direction. If this study uses static Gabor, random dot motion, object images that span from low-level to high-level visual stimuli, and consistently shows that the stimulus heterogeneity is encoded in one brain region, I would say this finding is valuable. But this sounds like another experiment. In other words, it is insufficient to claim a new brain region given the current form of the manuscript.
REFERENCES
- Duncan, J., & Humphreys, G. W. (1989). Visual search and stimulus similarity. Psychological Review, 96(3), 433-458. doi: 10.1037/0033-295x.96.3.433
- Duncan, J. (1989). Boundary conditions on parallel processing in human vision. Perception, 18(4), 457-469. doi: 10.1068/p180457
- Nagy, A. L., Neriani, K. E., & Young, T. L. (2005). Effects of target and distractor heterogeneity on search for a color target. Vision Research, 45(14), 1885-1899. doi: 10.1016/j.visres.2005.01.007
- Nagy, A. L., & Thomas, G. (2003). Distractor heterogeneity, attention, and color in visual search. Vision Research, 43(14), 1541-1552. doi: 10.1016/s0042-6989(03)00234-7
- Vincent, B., Baddeley, R., Troscianko, T., & Gilchrist, I. (2009). Optimal feature integration in visual search. Journal of Vision, 9(5), 15-15. doi: 10.1167/9.5.15
- Singh, A., Mihali, A., Chou, W. C., & Ma, W. J. (2023). A Computational Approach to Search in Visual Working Memory.
- Mihali, A., & Ma, W. J. (2020). The psychophysics of visual search with heterogeneous distractors. BioRxiv, 2020-08.
- Calder-Travis, J., & Ma, W. J. (2020). Explaining the effects of distractor statistics in visual search. Journal of Vision, 20(13), 11-11. -
-
-
eLife assessment
This study uses carefully designed experiments to generate a useful behavioural and neuroimaging dataset on visual cognition. The results provide solid evidence for the involvement of higher-order visual cortex in processing visual oddballs and asymmetry. However, the evidence provided for the very strong claims of homogeneity as a novel concept in vision science, separable from existing concepts such as target saliency, is inadequate.
-
Reviewer #1 (Public Review):
Summary:
The authors define a new metric for visual displays, derived from psychophysical response times, called visual homogeneity (VH). They attempt to show that VH is explanatory of response times across multiple visual tasks. They use fMRI to find visual cortex regions with VH-correlated activity. On this basis, they declare a new visual region in the human brain, area VH, whose purpose is to represent VH for the purpose of visual search and symmetry tasks.
Strengths:
The authors present carefully designed experiments, combining multiple types of visual judgments and multiple types of visual stimuli with concurrent fMRI measurements. This is a rich dataset with many possibilities for analysis and interpretation.
Weaknesses:
The datasets presented here should provide a rich basis for analysis. However, in …
Reviewer #1 (Public Review):
Summary:
The authors define a new metric for visual displays, derived from psychophysical response times, called visual homogeneity (VH). They attempt to show that VH is explanatory of response times across multiple visual tasks. They use fMRI to find visual cortex regions with VH-correlated activity. On this basis, they declare a new visual region in the human brain, area VH, whose purpose is to represent VH for the purpose of visual search and symmetry tasks.
Strengths:
The authors present carefully designed experiments, combining multiple types of visual judgments and multiple types of visual stimuli with concurrent fMRI measurements. This is a rich dataset with many possibilities for analysis and interpretation.
Weaknesses:
The datasets presented here should provide a rich basis for analysis. However, in this version of the manuscript, I believe that there are major problems with the logic underlying the authors' new theory of visual homogeneity (VH), with the specific methods they used to calculate VH, and with their interpretation of psychophysical results using these methods. These problems with the coherency of VH as a theoretical construct and metric value make it hard to interpret the fMRI results based on searchlight analysis of neural activity correlated with VH. In addition, the large regions of VH correlations identified in Experiments 1 and 2 vs. Experiments 3 and 4 are barely overlapping. This undermines the claim that VH is a universal quantity, represented in a newly discovered area of the visual cortex, that underlies a wide variety of visual tasks and functions.
Maybe I have missed something, or there is some flaw in my logic. But, absent that, I think the authors should radically reconsider their theory, analyses, and interpretations, in light of the detailed comments below, to make the best use of their extensive and valuable datasets combining behavior and fMRI. I think doing so could lead to a much more coherent and convincing paper, albeit possibly supporting less novel conclusions.
THEORY AND ANALYSIS OF VH
- VH is an unnecessary, complex proxy for response time and target-distractor similarity.
VH is defined as a novel visual quality, calculable for both arrays of objects (as studied in Experiments 1-3) and individual objects (as studied in Experiment 4). It is derived from a center-to-distance calculation in a perceptual space. That space in turn is derived from the multi-dimensional scaling of response times for target-distractor pairs in an oddball detection task (Experiments 1 and 2) or in a same-different task (Experiments 3 and 4). Proximity of objects in the space is inversely proportional to response times for arrays in which they were paired. These response times are higher for more similar objects. Hence, proximity is proportional to similarity. This is visible in Fig. 2B as the close clustering of complex, confusable animal shapes.
VH, i.e. distance-to-center, for target-present arrays, is calculated as shown in Fig. 1C, based on a point on the line connecting the target and distractors. The authors justify this idea with previous findings that responses to multiple stimuli are an average of responses to the constituent individual stimuli. The distance of the connecting line to the center is inversely proportional to the distance between the two stimuli in the pair, as shown in Fig. 2D. As a result, VH is inversely proportional to the distance between the stimuli and thus to stimulus similarity and response times. But this just makes VH a highly derived, unnecessarily complex proxy for target-distractor similarity and response time. The original response times on which the perceptual space is based are far more simple and direct measures of similarity for predicting response times.
- The use of VH derived from Experiment 1 to predict response times in Experiment 2 is circular and does not validate the VH theory.
The use of VH, a response time proxy, to predict response times in other, similar tasks, using the same stimuli, is circular. In effect, response times are being used to predict response times across two similar experiments using the same stimuli. Experiment 1 and the target present condition of Experiment 2 involve the same essential task of oddball detection. The results of Experiment 1 are converted into VH values as described above, and these are used to predict response times in Experiment 2 (Fig. 2F). Since VH is a derived proxy for response values in Experiment 1, this prediction is circular, and the observed correlation shows only consistency between two oddball detection tasks in two experiments using the same stimuli.
- The negative correlation of target-absent response times with VH as it is defined for target-absent arrays, based on the distance of a single stimulus from the center, is uninterpretable without understanding the effects of center-fitting. Most likely, center-fitting and the different VH metrics for target-absent trials produce an inverse correlation of VH with target-distractor similarity.
The construction of the VH perceptual space also involves fitting a "center" point such that distances to center predict response times as closely as possible. The effect of this fitting process on distance-to-center values for individual objects or clusters of objects is unknowable from what is presented here. These effects would depend on the residual errors after fitting response times with the connecting line distances. The center point location and its effects on the distance-to-center of single objects and object clusters are not discussed or reported here.
Yet, this uninterpretable distance-to-center of single objects is chosen as the metric for VH of target-absent displays (VHabsent). This is justified by the idea that arrays of a single stimulus will produce an average response equal to one stimulus of the same kind. However, it is not logically clear why response strength to a stimulus should be a metric for homogeneity of arrays constructed from that stimulus, or even what homogeneity could mean for a single stimulus from this set. It is not clear how this VHabsent metric based on single stimuli can be equated to the connecting line VH metric for stimulus pairs, i.e. VHpresent, or how both could be plotted on a single continuum.
It is clear, however, what *should* be correlated with difficulty and response time in the target-absent trials, and that is the complexity of the stimuli and the numerosity of similar distractors in the overall stimulus set. The complexity of the target, similarity with potential distractors, and the number of such similar distractors all make ruling out distractor presence more difficult. The correlation seen in Fig. 2G must reflect these kinds of effects, with higher response times for complex animal shapes with lots of similar distractors and lower response times for simpler round shapes with fewer similar distractors.
The example points in Fig. 2G seem to bear this out, with higher response times for the deer stimulus (complex, many close distractors in the Fig. 2B perceptual space) and lower response times for the coffee cup (simple, few close distractors in the perceptual space). While the meaning of the VH scale in Fig. 2G, and its relationship to the scale in Fig. 2F, are unknown, it seems like the Fig. 2G scale has an inverse relationship to stimulus complexity, in contrast to the expected positive relationship for Fig. 2F. This is presumably what creates the observed negative correlation in Fig. 2G.
Taken together, points 1-3 suggest that VHpresent and VHabsent are complex, unnecessary, and disconnected metrics for understanding target detection response times. The standard, simple explanation should stand. Task difficulty and response time in target detection tasks, in both present and absent trials, are positively correlated with target-distractor similarity.
I think my interpretations apply to Experiments 3 and 4 as well, although I find the analysis in Fig. 4 especially hard to understand. The VH space in this case is based on Experiment 3 oddball detection in a stimulus set that included both symmetric and asymmetric objects. However, the response times for a very different task in Experiment 4, a symmetric/asymmetric judgment, are plotted against the axes derived from Experiment 3 (Fig. 4F and 4G). It is not clear to me why a measure based on oddball detection that requires no use of symmetry information should be predictive of within-stimulus symmetry detection response times. If it is, that requires a theoretical explanation not provided here.
- Contrary to the VH theory, same/different tasks are unlikely to depend on a decision boundary in the middle of a similarity or homogeneity continuum.
The authors interpret the inverse relationship of response times with VHpresent and VHabsent, described above, as evidence for their theory. They hypothesize, in Fig. 1G, that VHpresent and VHabsent occupy a single scale, with maximum VHpresent falling at the same point as minimum VHabsent. This is not borne out by their analysis, since the VHpresent and VHabsent value scales are mainly overlapping, not only in Experiments 1 and 2 but also in Experiments 3 and 4. The authors dismiss this problem by saying that their analyses are a first pass that will require future refinement. Instead, the failure to conform to this basic part of the theory should be a red flag calling for revision of the theory.
The reason for this single scale is that the authors think of target detection as a boundary decision task, along a single scale, with a decision boundary somewhere in the middle, separating present and absent. This model makes sense for decision dimensions or spaces where there are two categories (right/left motion; cats vs. dogs), separated by an inherent boundary (equal left/right motion; training-defined cat/dog boundary). In these cases, there is less information near the boundary, leading to reduced speed/accuracy and producing a pattern like that shown in Fig. 1G.
This logic does not hold for target detection tasks. There is no inherent middle point boundary between target present and target absent. Instead, in both types of trials, maximum information is present when the target and distractors are most dissimilar, and minimum information is present when the target and distractors are most similar. The point of greatest similarity occurs at the limit of any metric for similarity. Correspondingly, there is no middle point dip in information that would produce greater difficulty and higher response times. Instead, task difficulty and response times increase monotonically with the similarity between targets and distractors, for both target present and target absent decisions. Thus, in Figs. 2F and 2G, response times appear to be highest for animals, which share the largest numbers of closely similar distractors.
DEFINITION OF AREA VH USING fMRI
- The area VH boundaries from different experiments are nearly completely non-overlapping.
In line with their theory that VH is a single continuum with a decision boundary somewhere in the middle, the authors use fMRI searchlight to find an area whose responses positively correlate with homogeneity, as calculated across all of their target present and target absent arrays. They report VH-correlated activity in regions anterior to LO. However, the VH defined by symmetry Experiments 3 and 4 (VHsymmetry) is substantially anterior to LO, while the VH defined by target detection Experiments 1 and 2 (VHdetection) is almost immediately adjacent to LO. Fig. S13 shows that VHsymmetry and VHdetection are nearly non-overlapping. This is a fundamental problem with the claim of discovering a new area that represents a new quantity that explains response times across multiple visual tasks. In addition, it is hard to understand why VHsymmetry does not show up in a straightforward subtraction between symmetric and asymmetric objects, which should show a clear difference in homogeneity.
- It is hard to understand how neural responses can be correlated with both VHpresent and VHabsent.
The main paper results for VHdetection are based on both target-present and target-absent trials, considered together. It is hard to interpret the observed correlations, since the VHpresent and VHabsent metrics are calculated in such different ways and have opposite correlations with target similarity, task difficulty, and response times (see above). It may be that one or the other dominates the observed correlations. It would be clarifying to analyze correlations for target-present and target-absent trials separately, to see if they are both positive and correlated with each other.
- The definition of the boundaries and purpose of a new visual area in the brain requires circumspection, abundant and convergent evidence, and careful controls.
Even if the VH metric, as defined and calculated by the authors here, is a meaningful quantity, it is a bold claim that a large cortical area just anterior to LO is devoted to calculating this metric as its major task. Vision involves much more than target detection and symmetry detection. The cortex anterior to LO is bound to perform a much wider range of visual functionalities. If the reported correlations can be clarified and supported, it would be more circumspect to treat them as one byproduct of unknown visual processing in the cortex anterior to LO, rather than treating them as the defining purpose for a large area of the visual cortex.
-
Reviewer #2 (Public Review):
Summary:
This study proposes visual homogeneity as a novel visual property that enables observers perform to several seemingly disparate visual tasks, such as finding an odd item, deciding if two items are the same, or judging if an object is symmetric. In Experiment 1, the reaction times on several objects were measured in human subjects. In Experiment 2, the visual homogeneity of each object was calculated based on the reaction time data. The visual homogeneity scores predicted reaction times. This value was also correlated with the BOLD signals in a specific region anterior to LO. Similar methods were used to analyze reaction time and fMRI data in a symmetry detection task. It is concluded that visual homogeneity is an important feature that enables observers to solve these two tasks.
Strengths:
- The …
Reviewer #2 (Public Review):
Summary:
This study proposes visual homogeneity as a novel visual property that enables observers perform to several seemingly disparate visual tasks, such as finding an odd item, deciding if two items are the same, or judging if an object is symmetric. In Experiment 1, the reaction times on several objects were measured in human subjects. In Experiment 2, the visual homogeneity of each object was calculated based on the reaction time data. The visual homogeneity scores predicted reaction times. This value was also correlated with the BOLD signals in a specific region anterior to LO. Similar methods were used to analyze reaction time and fMRI data in a symmetry detection task. It is concluded that visual homogeneity is an important feature that enables observers to solve these two tasks.
Strengths:
- The writing is very clear. The presentation of the study is informative.
- This study includes several behavioral and fMRI experiments. I appreciate the scientific rigor of the authors.
Weaknesses:
My main concern with this paper is the way visual homogeneity is computed. On page 10, lines 188-192, it says: "we then asked if there is any point in this multidimensional representation such that distances from this point to the target-present and target-absent response vectors can accurately predict the target-present and target-absent response times with a positive and negative correlation respectively (see Methods)". This is also true for the symmetry detection task. If I understand correctly, the reference point in this perceptual space was found by deliberating satisfying the negative and positive correlations in response times. And then on page 10, lines 200-205, it shows that the positive and negative correlations actually exist. This logic is confusing. The positive and negative correlations emerge only because this method is optimized to do so. It seems more reasonable to identify the reference point of this perceptual space independently, without using the reaction time data. Otherwise, the inference process sounds circular. A simple way is to just use the mean point of all objects in Exp 1, without any optimization towards reaction time data.
On page 11, lines 214-221. It says: "these findings are non-trivial for several reasons". However, the first reason is confusing. It is unclear to me why "it suggests that there are highly specific computations that can be performed on perceptual space to solve oddball tasks". In fact, these two sentences provide no specific explanation for the results.
The second reason is interesting. Reaction times in target-present trials can be easily explained by target-distractor similarity. But why does reaction time vary substantially across target-absent stimuli? One possible explanation is that the objects that are distant from the feature distribution elicit shorter reaction times. Here, all objects constitute a statistical distribution in the feature (perceptual) space. There is certainly a mean of this distribution. Some objects look like outliers and these outliers elicit shorter reaction times in the target-absent trials because outlier detection is very salient.
One might argue that the above account is merely a rephrasing of the idea of visual homogeneity proposed in this study. If so, feature saliency is not a new account. In other words, the idea of visual homogeneity is another way of reiterating the old feature saliency theory.
One way to reject the feature saliency theory is to compare the reaction times of the objects that are very different from other objects (i.e., no surrounding objects in the perceptual space, e.g., the wheel in the lower right corner of Fig. 2B) with the objects that are surrounded by several similar objects (e.g., the horse in the upper part of Fig. 2B). Also, please choose the two objects with similar distance from the reference point. I predict that the latter will elicit longer reaction times because they can be easily confounded by surrounding similar objects (i.e., four-legged horses can be easily confounded by four-legged dogs). If the density of object distribution per se influences the visual homogeneity score, I would say that the "visual homogeneity" is essentially another way of describing the distributional density of the perceptual space.
The searchlight analysis looks strange to me. One can easily perform a parametric modulation by setting visual homogeneity as the trial-by-trial parametric modulator and reaction times as a covariate. This parametric modulation produces a brain map with the correlation of every voxel in the brain. On page 17 lines 340-343, it is unclear to me what the "mean activation" is.
Minor points:
In the intro, it says: "using simple neural rules..." actually it is very confusing what "neural rules" are here. Better to change it to "computational principles" or "neural network models"??
In the intro, it says: "while machine vision algorithms are extremely successful in solving feature-based tasks like object categorization (Serre, 2019), they struggle to solve these generic tasks (Kim et al., 2018; Ricci et al. 2021). These are not generic tasks. They are just a specific type of visual task-judging relationship between multiple objects. Moreover, a large number of studies in machine vision have shown that DNNs are capable of solving these tasks and even more difficult tasks. Two survey papers are listed here.
Wu, Q., Teney, D., Wang, P., Shen, C., Dick, A., & Van Den Hengel, A. (2017). Visual question answering: A survey of methods and datasets. Computer Vision and Image Understanding, 163, 21-40.
Małkiński, M., & Mańdziuk, J. (2022). Deep Learning Methods for Abstract Visual Reasoning: A Survey on Raven's Progressive Matrices. arXiv preprint arXiv:2201.12382.
-
-
-
-
-