Do view-invariant brain representations of actions arise within 200 ms of viewing?

[R7I7]

Humans can rapidly visually recognise the actions people around them are engaged in and this ability is important for successful behaviour and social interaction. Isik et al. presented human subjects with 2-second video clips of humans performing actions while measuring brain activity with MEG. The clips comprised 5 actions (walk, run, jump, eat, drink) performed by each of five different actors and video-recorded from each of five different views (only frontal and profile used in MEG). Results show that action can be decoded from MEG signals arising about 200 ms after the onset of the video, with decoding accuracy peaking after about 500 ms and then decaying while the stimulus is still on, with a rebound after stimulus offset. Moreover, decoders generalise across actors and views. The authors conclude that the brain rapidly computes a representation that is invariant to view and actor.

ScreenShot738

Figure from the paper. Legend from the paper with my modifications in brackets: [Accuracy of action decoding (%) from MEG data as a function of time after video onset]. We can decode [which of five actions was being performed in the video clip] by training and testing on the same view (‘within-view’ condition), or, to test viewpoint invariance, training on one view (0 degrees [frontal, I think, but this should be clarified] or 90 degrees [profile]) and testing on the second view (‘across view’ condition). Results are each [sic] from the average of eight different subjects. Error bars represent standard deviation [across what?]. Horizontal line indicates chance decoding accuracy. […] Lines at the bottom of plot indicate significance with p<0.01 permutation test, with the thickness of the line indicating [for how many of the 8 subjects decoding was significant]. [Note the significant offset response after the 2-s video (whose duration should be indicated by a stimulus bar).]

 

The rapid view-invariant action decoding is really quite amazing. It would be good to see more detailed analyses to assess the nature of the signals enabling this decoding feat. Of course, 200 ms already allows for recurrent computations and the decodability peak is at 500 ms, so this is not strong evidence for a pure feedforward account.

The generalisation across actors is less surprising. This was a very controlled data set. Despite some variation in the appearance of the actors, it seems plausible that there would be some clustering of the vectors of pixels across space and time (or of features of a low-level visual representation) corresponding to different actors performing the same action seen from the same angle.

In separate experiments, the authors used static single frames taken from the videos and dynamic point-light figures as stimuli. These reduced form-only and motion-only stimuli were associated with diminished separation of actions in the human brain and in model representations, and with diminished human action recognition, suggesting that form and motion information are both essential to action recognition.

I’m wondering about the role of task-related priors. Subjects were performing an action recognition task on this controlled set of brief clips during MEG while freely viewing the clips (though this is not currently clearly stated). This task is likely to give rise to strong prior expectations about the stimulus (0 deg or 90 deg, one of five actions, known scale and positions of key features for action discrimination). Primed to attend to particular diagnostic features and to fixate in certain positions, the brain will configure itself for rapid dynamic discrimination among the five possible actions. The authors present a group-average analysis of eye movements, suggesting that these do not provide as much information about the actions as the MEG signal. However, the low-dimensional nature of the task space is in contrast to natural conditions, where a wider variety of actions can be observed and view, actor, size, and background vary more. The precise prior expectations might contribute to the rapid discriminability of the actions in the brain signals.

The authors model the results in the framework of feedforward processing in a Hubel-and-Wiesel/Poggio-style model that alternates convolution and max-pooling to simulate responses resembling simple and complex cells, respectively. This model is extended here to process video using spatiotemporal filter templates. The first layer uses Gabor filters, higher layers use templates in the first layer matching video clips in the stimulus set. The authors argue that this model supports invariant decoding and largely accounts for the MEG results.

Like the subjects, the model is set up to process the restricted stimulus space. The internal features of the model were constructed using representational fragments from samples from the same parametric space of videos. The exact videos used to test the models were not used for constructing the feature set. However, if I understood correctly, videos from the same restricted space (5 actions, 5 actors, 5 views) were used. Whether the model can be taken to explain (at a high level of abstraction) the computations performed by the brain depends critically on the degree to which the model is not specifically constructed for the (necessarily very limited) 5-action controlled stimulus space used in the study.

As the authors note, humans still outperform computer vision models at action recognition. How does the authors’ own model perform on less controlled action videos? If it the model cannot perform the task on real-world sensory input, can we be confident that it captures the way that the human brain performs the task? This is a concern in many studies and not trivial to address. However, the interpretation of the results should engage this issue.

 

Strengths

  • Controlled stimulus set: The set of video stimuli (5 actions x 5 actors x 5 views x 26 clips = 3250 2-sec clips) is impressive. Assembling this set is an important contribution to the field. The set is condition-rich (compared to typical stimulus sets used in cognitive neuroscience studies) and seems to strike a good balance between control and realism. This set could be a driver of progress if it were to be used in multiple modelling and empirical studies.
  • Combination of brain-activity measurements and a simple computational model, which provides a useful starting point for modelling the recognition of dynamic actions, as it is minimal and standard in many respects: a feedforward model in the HMAX framework, extended from spatial to spatiotemporal filters.

 

Weaknesses

  • Controlled stimulus set: The set of video stimuli is very restricted compared to real-world action recognition. For the brain data, this means that subjects might have rapidly formed priors about the stimuli, enabling them to configure their visual systems (e.g. attentional templates, fixation targets) for more rapid recognition of the 5 actions than is possible in real-world settings. This limitation is shared with many studies in our field and difficult to overcome without giving up control (which is a strength, see above). I therefore suggest addressing this problem in the discussion.
  • The model uses features based on spatiotemporal patterns sampled from the same restricted stimulus space. Although non-identical clips were used, the videos underlying the representational space appear to share a lot with the experimental stimuli (same 5 actions, same 5 views, same background?, same actors?). I would therefore not expect this model to work well on arbitrary real-world action video clips. This is in contrast to recent studies using deep convolutional neural nets (e.g. Khaligh-Razavi & Kriegeskorte 2014), where the models were trained without any information about the (necessarily restricted) brain-experimental stimulus set and can perform recognition under real-world conditions.
  • Only one model (in two variants) is tested. In order to learn about computational mechanism, it would be good to test more models.
  • MEG data were acquired during viewing of only 50 of the clips (5 actions x 5 actors x 2 views).
  • Missing inferential analyses: While the authors employ inferential analyses in single subjects and report number of significant subjects, few hypotheses of interest are formally statistically tested. The effects interpreted appear quite strong, so the results described above appear solid nevertheless (interpretational caveats notwithstanding).

 

Overall evaluation

This is an ambitious study describing results of a well-designed experiment using a stimulus set that is a major step forward. The results are an interesting and substantial contribution to the literature. However, the analyses could be much richer than they currently are and the interpretation of the results is not straightforward. Stimulus-set-induced priors may have affected both the neural processing measured and the model (which used templates from stimuli within the controlled video set). Results should be interpreted more cautiously in this context.

Although feedforward processing is an important part of the story, it is not the whole story. Recurrent signal flow is ubiquitous in the brain and essential to brain function. In engineering, similarly, recurrent neural networks are beginning to dominate spatiotemporal processing challenges such as speech and video recognition. The fact that the MEG data are presented as time courses, revealing a rich temporal structure, and the model analyses are bar graphs illustrates the key limitation of the model.

It would be great to extend the analyses to reveal a richer picture of the temporal dynamics. This should include an analysis of the extent to which each model layer can explain the representational geometry at each latency from stimulus onset.

 

Future directions

In revision or future studies, this line of work could be extended in a number of ways:

  • Use multiple models that can handle real-world action videos. The authors’ controlled video set is extremely useful for testing human and model representations, and for comparing humans to models. However, to be able to draw strong conclusions, the models, like the humans, would have to be trained to recognise human actions under real-world conditions (unrestricted natural video). In addition, it would be good to compare the biological representational dynamics to both feedforward and recurrent computational models.
  • To overcome the problem of stimulus-set related priors, which make it difficult to compare representational dynamics measured for restricted stimulus sets to real-world recognition in biological brains, one could present a large set of stimuli without ever presenting a stimulus twice to the same subject. Would the action category still be decodable at 150 ms with generalisation across views? Would a feedforward computer vision model trained on real-world action videos be able to predict the representational dynamics?
  • The MEG analyses could use source reconstruction to enable separate analyses of the representational dynamics in different brain regions.
  • It would be useful to have MEG data for the full stimulus set of 5 actions x 5 actors x 5 viewpoints = 125 conditions. The representational geometries could be analysed in detail to reveal which particular action pairs become discriminable when with what level of invariance.

 

 

Particular suggestions for improvements of this paper

(1) Present more detailed results

It would be good to see results separately for each pair of actions and each direction of crossdecoding (0 deg training -> 90 deg testing, and 90 deg training -> 0 deg testing). Regarding the former, eating and drinking involve very similar body postures and motions. Is this reflected in the discriminability of these actions?

Regarding, the decoding generalisation across views, you state:

“We decoded by training only on one view (0 degrees or 90 degrees), and testing on a second view (0 degrees or 90 degrees).”

Was the training set exclusively composed on 0 degree (frontal?) and the test set exclusively of 90 degree (side view?), and vice versa? In case the test set contained instances of both views (though of course, not for the same actor and action), results are more difficult to interpret.

 

(2) Discuss the caveats to the current interpretation of the results

Discuss the question whether priors resulting from subjects understanding of the restricted stimulus set might have affected the processing of the stimuli. Consider the involvement of recurrent computations within 200 ms and discuss the continuing rise of decodability until 500 ms. Discuss the possibility that the model will not generalise to action recognition in the wild.

 

(3) Test several control models

Can Gabor, HMAX, and deep convolutional neural net models support similarly invariant action decoding? These models are relatively easy to test, so I think it’s worth considering this for revision. Computer vision models trained on dynamic action recognition could be left to future studies.

 

(4) Test models by comparison of its representations with the brain representations

The computational model is currently only compared to the data at the very abstract level of decoding accuracy. Can the model predict the representations and representational dynamics in detail? It might be difficult to use the model to predict the measured channels. This would require the fitting of a linear model predicting the measured channels from the model units and the MEG data (acquired for only 5 actions x 5 actors x 2 views = 50 conditions) might be insufficient. However, representational dynamics could be investigated in the framework of representational similarity analysis (50 x 50 representational dissimilarity matrices) following Carlson et al. (2013) and Cichy et al. (2014). Note that this approach does not require fitting a prediction model and so appears applicable here. Either approach would reveal the dynamic prediction of the feedforward model (given dynamic inputs) and where its prediction diverges from the more complex and recurrent processes in the brain. This would promise to give us a richer and less purely confirmatory picture of the data and might show the merits and limitations of a feedforward account.

 

(5) Perform temporal cross-decoding

Temporal crossdecoding (Carlson et al. 2013, Cichy et al. 2014) could be used to more richly characterise the representational dynamics. This would reveal whether representations stabilise in certain time windows, or keep tumbling through the representational space even as stimuli are continuously decodable.

 

(6) Improve the inferential analyses

I don’t really understand the inference procedure in detail from the description in the methods section.

“We recorded the peak decoding accuracy for each time bin,…”

What is the peak decoding accuracy for each time bin? Is this the maximum accuracy across subjects for each time bin?

“…and used the null distribution of peak accuracies to select a threshold where decoding results performing above all points in the null distribution for the corresponding time point were deemed significant with P < 0.01 (1/100).”

I’m confused after reading this, because I don’t understand what is meant by “peak”.

The inference procedure for the decoding-accuracy time courses seems to lack formal multiple-testing correction across time points. Given enough subjects, inference could be performed with subject as a random effect. Alternatively, fixed-effects inference could be performed by permutation, averaging across subjects. Multiple testing across latencies should be formally corrected for. A simple way to do this is to relabel the experimental events once, compute an entire decoding time course, and record the peak decoding accuracy across time (or if this is what was done, it should be clearly described). Through repeated permutations, a null distribution of peak accuracies can be constructed and a threshold selected that is exceeded anywhere under H0 with only 5% probability, thus controlling the familywise error rate at 5%. This threshold could be shown as a line or as the upper edge of a transparent rectangle that partially obscures the insignificant part of the curve.

For each inferential analysis, please describe exactly what the null hypothesis was, what event-labels are exchangeable under this null hypothesis, and how the null distribution was computed. Also, explain how the permutation test interacted with the crossvalidation procedure. The crossvalidation should ideally generalise to new stimuli and label permutation be wrapped around this entire procedure.

“Decoding analysis was performed using cross validation, where the classifier was trained on a randomly selected subset of 80% of data for each stimulus and tested on the held out 20%, to assess the classifier’s decoding accuracy.”

Does this apply only to the within-view decoding? In the critical decoding analysis with generalisation across views, it cannot have been 20% of the data in the held-out set, since 0-deg views were used for training and 90-deg views for testing (and vice versa). If only 50% of the data were used for training there, why didn’t performance suffer given the smaller training set compared to the within-view decoding?

It would also be good to have estimates and inferential comparisons of the onset and peak latencies of the decoding time courses. Inference could be performed on a set of single-subject latency differences between two conditions modelling subject as a random effect.

 

(7) Qualify claims about biological fidelity of the model

The model is not really “designed to closely mimic the biology of the visual system”, rather its architecture is inspired by some of the features of the feedforward component of the visual hierarchy, such as local receptive fields of increasing size across a hierarchy of representations.

 

(8) Open stimuli and data

This work would be especially useful to the community if the video stimuli and the MEG data were made openly available. To fully interpret the findings, it would also be good to be able to view the movie clips online.

 

(9) Further clarify the title

The title “Fast, invariant representation for human action in the visual system” is somewhat unclear. What is meant are representations of perceived human actions, not representations for action. “Fast, invariant representation of visually perceived human actions” would be better, for example.

 

(10) Clarify what stimuli MEG data were acquired for

The abstract states “We use magnetoencephalography (MEG) decoding and a computational model to study action recognition from a novel dataset of well-controlled, naturalistic videos of five actions (run, walk, jump, eat, drink) performed by five actors at five viewpoints.” This suggests that MEG data were acquired for all these conditions. The methods section clarifies that MEG data were only recorded for 50 conditions (5 actions x 5 actors x 2 views). Here and in the legend of Fig. 1, it would be better to use the term “stimulus set” in place of “data set”.

 

(11) Clarify whether subjects were fixating or free viewing

Were subjects free viewing or fixating? This should be explicitly stated and the choice motivated in either case.

 

(12) Make figures more accessible

The figures are not optimal. Every panel showing decoding results should be clearly labelled to state what variables the crossvalidation procedure tested for generalisations across. For example, a label (in the figure itself!) could be: “decoding brain representations of actions with invariance to actor and view”. The reader shouldn’t have to search in the legend to find this essential information. Also every figure should have a stimulus bar depicting the period of stimulus presence. This is important especially to assess stimulus-offset-related effects, which appear to be present and significant.

Fig. 3 is great. I think it would be clearer to replace “space-time dot product” with “space-time convolution”.

 

(13) Clarify what the error bars represent

“Error bars represent standard deviation.”

Is this the standard deviation across the 8 subjects? Is it really the standard deviation or the standard error?

 

 (14) Clarify what we learn from the comparison between the structured and the unstructured model

For the unstructured model, won’t the machine learning classifier learn to select random combinations that tend to pool across different views of one action? This would render the resulting system computationally similar.

 

 

 

Pattern-component modelling disentangles the code and the noise in representational similarity analysis

[R8I8]

This paper proposes an interesting and potentially important extension to representational similarity analysis (RSA), which promises unbiased estimates of response-pattern similarities and more compelling comparisons of representations between different brain regions.

RSA consists in the analysis of the similarity structure of the representations of different stimuli (or mental states associated with different tasks) in a region of interest (ROI). To this end, the similarity of regional response patterns elicited by the different stimuli is estimated, typically by using their linear correlation coefficient across voxels (or neurons or recording sites in electrophysiology). It is often desirable to be able to compare these pattern similarities between different regions. For example, we would like to be able to address whether stimuli A and B elicit more highly correlated response patterns in region 1 or region 2. However, such comparisons are problematic, because the pattern correlations depend on fMRI noise (which might be different between the regions), voxel selection (e.g. selecting additional noisy voxels will reduce the pattern correlation), and unspecific pattern components (e.g. a strong shared component between all stimuli will increase the pattern correlation, with the high correlation not specific to the particular pair of stimuli).

ScreenShot687
Pattern-component modelling yields estimates of the similarity of representational patterns that are not systematically distorted by noise and common components. Representational pattern similarity is measured here by the correlation across measurement channels (e.g. fMRI voxels) and is plotted as a function of the noise level (horizontal axes) for different amplitudes (shades of gray) of a common pattern component shared by both representational patterns. Figure from Diedrichsen et al. (2011).

When representational dissimilarities (or, equivalently similarities) are estimated from estimates of response patterns in a multidimensional space, the dissimilarity estimates are positively (or the similarity estimates negatively) biased. This is because the inevitable noise affecting the pattern estimates will typically increase the apparent distance between any two patterns (the probability of a decrease of the distance due to noise is 0.5 in 1 dimension and drops rapidly as dimensionality increases).

Instead of estimating the distances from pattern estimates, the authors therefore propose to estimate the distances from a covariance component model that captures the pattern variances and covariances across space. The approach requires that each stimulus (or, more generally, each experimental condition) has been repeated multiple times to yield multiple pattern estimates. Whereas simple RSA would consider the average pattern for each stimulus, the authors’ approach models the original trial-by-voxel matrix Y as a linear combination of a set of stimulus-related patterns U (thought to underlie the observed patterns) and  noise, and estimates the covariance structure of the patterns. The noise E is assumed to be independent between trials, but there is no assumption of independence of the noise between voxels. This is important because fMRI error time series from voxels closeby within a region are known to be correlated.

This is an original and potentially important contribution. The core mathematical model appears well developed. The demonstration of the advantages of the method is compellingly demonstrated based on simulated data. The paper is well written. However, it requires a number of improvements to ensure that it successfully communicates its important message. (1) The authors should more clearly explain the assumptions their pattern-covariance modelling approach relies upon. (2) The authors should add a section explaining the practical application of the approach (3) A number of clarifications and didactical improvements, notably to the presentation of the analysis of the real fMRI data, would be desirable. These three major points are explained in detail below.

[This is my original secret peer review of Diedrichsen et al. (2011). Most of the suggestions for improvements were addressed in revision and are no longer relevant.]

MAJOR POINTS

(1) Assumptions and consequences of violations

The advantages of pattern-covariance modeling are well explained. However, the assumptions of this approach should be more clearly communicated, perhaps in a separate section.

  • Does the validity of the approach depend on assumptions about the probability densities of the response amplitudes? Are there any other assumptions about the nature of the response patterns?
  • What are the effects of violations of the assumptions? Please give examples of cases where the assumptions are violated and describe the expected effects on the analysis.
  • As long as statistical inference is performed at the level of the variability across subjects or by using randomisation testing, results might be robust to certain violations. Please clarify if and when this is the case.

 

(2) Practical application of the new approach

Please add a section explaining how to apply this method to fMRI data, addressing the following questions:

  • Do the authors plan to make matlab code available for the new method? If so, it would be good to state this in the paper.
  • Is there a function that takes the regional data matrix Y, the design matrix Z (including effects of no interest) and perhaps a predictor selection vector for selecting effects of interest as input and returns the corrected correlation (and perhaps Euclidean) distance matrix?
  • Does the method only work with slow event-related designs (with approximately independent trial estimates)?
  • Can we use the method on rapid-event-related designs where we do not have separate single-trial estimates (because single-trial responses overlap in time and multiple trials of the same condition must be estimated together for stability)?
  • What if we have only one pattern estimate per condition, because our design is condition-rich (e.g. 96 conditions as in Kriegeskorte et al. 2008) and rapid-event related?
  • More generally, what are the requirements and limitations of the proposed approach?

 

(3) Particular clarifications and didactical improvements

In classical multivariate regression, we get an estimate of the error of a spatial response pattern estimate as a multinormal (characterised by a scaled version of the voxel-by-voxel covariance matrix of the residuals, where the scaling factor reflects the amount of averaging for the case of binary nonoverlapping predictors, and, more generally, the sums of squares and products of the design matrix). Couldn’t this multinormal model of the variability of each condition-related pattern estimate be used to get an unbiased estimate of the correlation of each pair of pattern estimates? If so, would this approach be inferior or superior to the proposed method, and why?

  1. 7: What exactly are the ‘simplifying assumptions’ that allow a to be estimated independently of G by averaging the trial response patterns within conditions?

“The corrected estimate from the covariance-component model is unbiased over a large range of parameter settings.” What are the limits of this range? Is the estimate formally unbiased or just approximately so?

Can question a) “Does the region encode information about the finger in the movement and/or stimulation condition?” be addressed with the traditional and the proposed RSA method? It seems that that would necessitate estimating the replicability of patterns elicited by moving the same finger (and similarly for sensation). It is a typical and important neuroscientific question, so please consider addressing in the framework of RSA (not just in terms of a possible classifier analysis as in the current draft).

Across different runs, pattern correlations are usually found to be much lower (e.g. Misaki et al. 2010). This phenomenon requires further investigation. The authors suggest error correlations among trials closeby in time within a run as the cause. However, I suspect that such error correlations, though clearly present, might not be the major cause of this. Other causes include scanner drifts and greater head-motion-related disalignment (due to greater separation in time), which can cause distortions, that head-motion-correction cannot undo. It would be good to hear the authors’ assessment of these alternative causes.

The notation u_beta[c,1,…4], where c is an element of {1,2} is confusing to me. Shouldn’t it be u_beta[c,d], where c is an element of {1,2}, and d is an element of {1,2,3,4}?

Eq. 8 requires more unpacking. Perhaps a figure with the vertical and horizontal dimensions marked (“task effects: movement vs sensation”, “individual finger effects: (1) movement, (2) sensation”) and arrows pointing from conceptual labels (“shared pattern between all movement trials”, “shared pattern between all sensation trials”, etc.) to the variance components could serve this function.

Figures 1-4 are great.

Figures 6 and 7: This comparison between traditional RSA and the proposed method is not completely successful. Figure 6 the traditional approach is very comprehensible. Figure 7 is cryptic (partly due to lack of meaningful labeling of the vertical axes). Moreover, the relationship between the traditional and the proposed approach to RSA remains unclear (or anyway difficult to grasp at a glance). I suggest adding a figure that compares traditional RSA and the proposed method side by side. The top row should show the correlation matrices (sample correlation versus unbiased estimates from covariance component model). The next three rows should address the three questions raised in the text: “a) Does the region encode information about the finger in the movement and/or stimulation condition? b) Are the patterns evoked by movement of a given finger similar to the patterns evoked by stimulation of the same finger? c) Is this similarity greater in one region than another?” Results from the traditional and the proposed RSA should be shown for each question to demonstrate how the results appear in both approaches and where the traditional approach falls short.

 

 

MINOR POINTS

In Eq. 8, u_beta[1,2] should read u_beta[1,1], I think.

“The decomposition method offers an elegant way to control for all these possible influences on the size of the correlation coefficients. In addition to noise (ε), condition ( , ), and finger ( , ) effects (Eq. 7), we also added a run effect.” Should say Eq. 8, I think.

Does U stand for ‘(u)nderlying patters’ and a for spatial-average (a)ctivation? It would help to make this explicit.

Figure 6 : Please label the vertical axes (intuitive and clear conceptual label). Please mark all significant effects. Please add a colorbar (grayscale code for correlation). Legend: “(D) These correlations” Which correlations exactly? (Averaged across sense and move now?)

Figure 7: The vertical axes need to be intuitively labeled. The reader should not have to decode mathematical symbols from the legend to understand the meaning of the bar graphs. Even after a careful read of the legend (and after spending quite a bit of time on the paper), the neuroscientific findings are not easy to grasp here. As a result, the present version of this figure will leave readers preferring traditional RSA (Figure 6) as it at least can be interpreted without much effort. Please label gray and white (“sense” and “move”) bars as in Figure 6.

 

 

 

 

Imagining and seeing objects elicits consistent category-average activity patterns in the ventral stream

[R8I7]

Horikawa and Kamitani report results of a conceptually beautiful and technically sophisticated study decoding the category of imagined objects. They trained linear models to decode visual image features from fMRI voxel patterns. The visual features are computed from images by computational models including GIST and the AlexNet deep convolutional neural net. AlexNet provides features spanning the range from visual to semantic. A subject is then scanned while imagining images from a novel object category (not used in training the fMRI decoder). The decoder is used to predict the computational-model representation for the imagined category (averaged across exemplars of that category). This predicted model representation is then compared to the actual model representation for many categories, including the imagined one. The model representation predicted from fMRI during imagery is shown to be significantly more similar to the model representation of images from the imagined category than to the model representation of images from other categories.

ScreenShot685

Figure from Horikawa & Kamitani (2015)

The methods are sophisticated and will give experts much to think about and draw from in developing better decoders. Comprehensive supplementary analyses, which I did not have time to fully review, complement and extend the thorough analyses provided. This is a great study. As usual in our field, a difficult question is what exactly it means for brain computational theory.

A few results that might speak to the computational mechanism of the ventral stream are as follows.

When predicting computational features of *single images* (which was only done for seen, not for imagined objects):

  • Lower layers of AlexNet are better predicted from voxels in lower ventral-stream areas.
  • Higher layers of AlexNet are better predicted from voxels in higher ventral-stream areas.
  • GIST features are best predicted from V1-3, but also significantly from higher areas.

This is consistent with the recent findings (Yamins, Khaligh-Razavi, Cadieu, Guclu) showing that deep convolutional neural nets explain lower- and higher-level ventral-stream areas with a rough correspondence of lower model layers to lower brain areas and higher model layers to higher brain areas. It is also consistent with previous findings that GIST, like many visual feature models, explains significant representational variance even in the higher ventral-stream representation (Khaligh-Razavi, Rice), but does not reach the noise ceiling (indicating that a data set is fully explained), as deep neural net models do (Khaligh-Razavi).

When predicting *category-averages* of computational features (which was done for seen and imagined objects):

  • Higher-level visual areas better predict features in all layers of AlexNet.
  • Higher layers of AlexNet are better predicted from voxels in all visual areas.

This is confusing, until we remember that it is category averages that are being predicted. Category averaging will retain a major portion of the representational variance of category-sensitive higher-level representations, while reducing the representational variance of low-level representations that are less related to categories. This may boost both predictions from category-related visual areas, as well as predictions of category-related model features.

Subjects imagined many different images from a given category in an experimental block during fMRI. The category-average imagery activity of the voxels was then used to predict the corresponding category-averages of the computational-model features. As expected, category-average computational-feature prediction is worse for mental imagery than for perception. The pattern across visual areas and AlexNet layers is similar for imagery and perception, with higher predictions resulting when the predicting visual area is category-related and when the predicted model feature is category-related. However, V1 and V2 did not consistently enable imagery decoding into the format of any of the layers of AlexNet. Interestingly, computational features more related to categories were better decodable. This supports the view that higher ventral-stream features might be optimised to emphasise categorical divisions (cf Jozwik et al. 2015).

 

Suggested improvements

(1) Clarify any evidence about the representational format in which the imagined content is represented. The authors’ model predicts both visual and semantic features of imagined object categories. This suggests that imagery involves both semantic and visual representations. However, the evidence for lower- or even mid-level visual representation of imagined objects is not very compelling here, because the imagery was not restricted to particular images. Instead the category-average imagery activity was measured. Each category is, of course, associated with particular visual features to some extent. We therefore expect to be able to predict category-average visual features from category-average voxel patterns better than chance. A strong claim that imagery paints low-level visual features into early visual representations would require imagery of particular images within each category. For relevant evidence, see Naselaris et al. (2015).

(2) Go beyond the decoding spin: what do we learn about computations in the ventral stream? Being able to decode brain representations is cool because it demonstrates unambiguously that a certain kind of information is present in a brain region. It’s even cooler to be able to decode into an open space of features or categories and to decode internally generated representations as done here. Nevertheless, the approach of decoding is also scientifically limiting. From the present version of the paper, the message I take is summarised in the title of the review: “Imagining and seeing objects elicits consistent category-average activity patterns in the ventral stream”. This has been shown previously (e.g. Stokes, Lee), but is greatly generalised here and is a finding so important that it is good to have it replicated and generalised in multiple studies. The reason why I can’t currently take a stronger computational claim from the paper is that we already know that category-related activity patterns cluster hierarchically in the ventral stream (Kriegeskorte et al. 2008) and may be continuously and smoothly related to a semantic space (Mitchell et al. 2008; Huth et al. 2012). In the context of these two pieces of knowledge, consistent category-average activity for perception and imagery is all that is needed to explain the present findings of decodability of novel imagined categories. The challenge to the authors: Can you test specific computational hypotheses and show something more on the basis of this impressive experiment? The semantic space analysis goes in this direction, but did not appear to me to support totally novel theoretical conclusions.

(3) Why decode computational features? Decoding of imagined content could be achieved either by predicting measured activity patterns from model representations of the stimuli (e.g. Kay et al. 2008) or by predicting model representations  from measured activity patterns (the present approach). The former approach is motivated by the idea that the model should predict the data and lends itself to comparing multiple models, thus contributing to computational theory. We will see below that the latter approach (chosen here) is less well suited to comparing alternative computational models. Why did Horikawa & Kamitani choose this approach? One argument might be that there are many model features and predicting the smaller number of voxels from these many features requires strong prior assumptions (implicit to regularisation), which might be questionable. The reverse prediction from voxels to features requires estimating the same total number of weights (# voxels * # model features), but each univariate linear model predicting a feature only has # voxels (i.e. typically fewer than # features) weights. Is this why you preferred this approach? Does it outperform the voxel-RF modelling approach of Kay et al. (2008) for decoding?

An even more important question is what we can learn about brain computations from feature decoding. If V4, say, perfectly predicted CNN1, this would suggest that V4 contains features similar to those in CNN1. However, it might additionally contain more complex features unrelated to CNN1. CNN1 predictability from V4, thus, would not imply that CNN1 can account for V4. Another example: CNN8 and GIST features are similarly predictable from voxel data across brain areas, and most predictable from V4 voxels. Does this mean GIST is as good a model as CNN8 for explaining the computational mechanism of the ventral stream? No. Even if the ventral-stream voxels perfectly predicted GIST, this would not imply that GIST perfectly predicts the ventral-stream voxels.

The important theoretical question is what computational mechanism gives rise to the representation in each area. For the human inferior temporal cortex, Khaligh-Razavi & Kriegeskorte (2015) showed that both GIST and the CNN representation explain significant variance. However, the GIST representation leaves a large portion of the explainable variance unexplained, whereas the CNN fully explains the explainable variance.

(4) Further explore the nature of the semantic space. To understand what drives the decoding of imagined categories, it would be helpful to see the performance of simpler analyses. Following Mitchell et al. (2008), one could use a text-corpus based semantic embedding to represent each of the categories. Decoding into this semantic embedding would similarly enable novel seen and imagined test categories (not used in training) to be decoded. It would be interesting, then, to successively reduce the dimensionality of the semantic embedding to estimate the complexity of the semantic space underlying the decoding. Alternatively, the authors’ WordNet distance could be used for decoding.

(5) Clarify that category-average patterns were used. The terms “image-based information” and “object-based information” are not ideal. By “image-based”, you are referring to a low-level visual representation and by “object-based”, to a categorical representation. Similarly, in many places where you say “objects” (as in “decoding objects”) it would be clearer to say “object categories”. Use clearer language throughout to clarify when it was category-average patterns that were used for prediction (brain representations) and that were predicted (model representations). This concerns the text and the figures. For example, the title of Fig. 4 should be: “Object-category-average feature decoding”. If this detracts the casual reader too lazy to even read the legends too much, at least the text of the legend should clearly state that category-average brain activity patterns are used to predict category-average model features.

(6) What are the assumptions implicit to sparse linear regression and is this approach optimal? L2 regularisation would spread the weights out over more voxels and might benefit from averaging out the noise component. Please comment on this choice and on any alternative performance results you may have.

 

Minor points

(7) The work is related to Mitchell et al. (2008), who predicted semantic semantic brain representations of novel stimuli using a semantic space model. This paper should be cited.

(8) “These studies showed a high representational similarity between the top layer of a convolutional neural network and visual cortical activity in the inferior temporal (IT) cortex of humans [24,25] and non-human primates [22,23].”

Ref 24 showed this for both human fMRI and macaque cell-recording data.

(9) “Interestingly, mid-level features were the most useful in identifying object categories, suggesting the significant contributions of mid-level representations in accurate object identification.”

This sentence repeats the same point after “suggesting”.