Claims 34
From the paper Perceptual Learning Reduces Interneuronal Correlations in Macaque Visual Cortex · claims file · checked 2026-11-06: ready
- fig2b-trained-lower-noise-correlation
Mean noise correlation among MSTd pairs (separated by less than 1 mm) is significantly smaller in trained animals (0.023 +- 0.017 SEM, n = 89) than in naive animals (0.116 +- 0.031, n = 38).
The mean noise correlation in the trained group (0.023 ± 0.017 SEM, n = 89) was significantly smaller than that for naive animals (0.116 ± 0.031, n = 38, p = 0.006, t test)
Results, Comparison of Noise Correlations in Trained and Naive Animals, paragraph 2; Figure 2BHow it will be judged
rnoise per pair as in the Experimental Procedures (Noise and Signal Correlations), pooled across headings and across the visual and vestibular conditions, for the 127 pairs under 1 mm. Tolerances on the means are the reported SEMs. The paper names only "t test"; the readings are a two-sided two-sample Student t test and a Welch t test. Whether the reported values include the renormalization in blocks of 20 trials is open; both readings are run. Reproduced when, under one reading, the trained mean is below the naive mean with p < 0.05 and both means are within tolerance; partially reproduced when the difference is significant but a mean falls outside tolerance.
- fig2c-timecourse-naive-higher
Noise correlation is greater in naive than in trained animals throughout the time course of the response, computed in 500 ms windows stepped by 50 ms (p = 0.002, permutation test).
rnoise was significantly greater in naive than trained animals throughout the time course of the neural response (p = 0.002, permutation test, see Experimental Procedures)
Results, Comparison of Noise Correlations in Trained and Naive Animals, paragraph 3; Figure 2CHow it will be judged
Needs spike times. rnoise per pair and window as for the main measure, in 500 ms windows moved in 50 ms steps (31 windows), averaged per group; the permutation test of the Experimental Procedures (sum of squared differences of the two group time courses, pairs drawn at random from the pooled groups into groups of the original sizes, 10,000 permutations, p the proportion of permuted values above the observed one). Where the 31 windows sit relative to stimulus onset is not stated; the reading is windows starting at stimulus onset and ending at its 2 s end (0-500 ms to 1500-2000 ms), which gives exactly 31. Reproduced when p < 0.05 and the naive mean is above the trained mean in every window; partially reproduced when p < 0.05 and the naive mean is higher in most windows.
- figs2a-difference-decreases
The difference in rnoise between naive and trained animals is largest at the beginning of the trial and decreases with time (Spearman R = -0.9, p << 0.001).
The difference in rnoise between naive and trained animals was largest at the beginning of the trial and gradually decreased with time
Results, Comparison of Noise Correlations in Trained and Naive Animals, paragraph 3; Figure S2AHow it will be judged
The 31 window differences (naive mean minus trained mean) of fig2c, correlated with window time by Spearman rank correlation. Reproduced when R is within tolerance and p < 0.001; partially reproduced when R is negative with p < 0.05.
- figs2b-individual-animals
The greater noise correlation in naive animals holds for individual animals; four trained animals (35, 26, 18 and 8 pairs) and three naive animals (22, 11 and 5 pairs) are shown, and the difference is consistent across animals within each group.
Importantly, these observations held true when correlations were examined for individual animals
Results, Comparison of Noise Correlations in Trained and Naive Animals, paragraph 3; Figure S2B; Discussion, paragraph 1How it will be judged
The paper shows per-animal time courses and gives no test, so the comparison is a reading of the figure and the Discussion sentence that the difference "was highly significant and consistent across animals within each group". Per animal, the mean rnoise of its pairs (pooled measure, middle 1 s). Pair counts per animal are checked against 35, 26, 18, 8 (and 2, not shown) trained and 22, 11, 5 naive. Reproduced when every naive animal's mean exceeds every trained animal's mean among the seven animals shown; partially reproduced when every naive animal's mean exceeds the trained group mean. Time courses per animal are shown when spike times are available.
- fig1-example-pair
For the example pair of Figure 1, signal correlation is 0.83 (visual) and 0.79 (vestibular), and noise correlation is 0.29 (visual, p = 0.04) and 0.14 (vestibular, p = 0.3).
For this example pair of neurons, rsignal = 0.83 and 0.79 for the visual and vestibular conditions, respectively ... rnoise = 0.29, p = 0.04, Figure 1D ... R = 0.14, p = 0.3, Figure 1E
Results, paragraphs 1 and 2; Figure 1B to 1EHow it will be judged
The paper does not identify the pair. Reproduced when one pair in the data has all four values within tolerance; not reproduced when none does.
- fig1f-modality-means-equal
Mean noise correlation does not differ between the vestibular (0.035 +- 0.014 SEM) and visual (0.039 +- 0.015) conditions (p > 0.8, paired t test).
their means were not significantly different (vestibular ... 0.035 ± 0.014 SEM, visual ... 0.039 ± 0.015, p > 0.8, paired t test)
Results, Noise Correlations in Area MSTd, paragraph 2; Figure 1FHow it will be judged
A claim of absence, judged on the paper's test, a paired t test at p < 0.05 on the per-condition rnoise of the 179 pairs (the 127 pairs under 1 mm when only those are available, noted). Reproduced when the test finds no difference; the means and the difference are reported either way.
- figs1-distance
Noise correlations depend on distance; mean rnoise is above zero for pairs under 0.05 mm (0.042 +- 0.021, p = 0.049, n = 67) and 0.05 to 1 mm (0.062 +- 0.024, p = 0.011, n = 60), but not for pairs over 1 mm (0 +- 0.15, p = 0.9, n = 52).
Average noise correlations were significantly greater than zero for the first two groups (<0.05 mm ... 0.042 ± 0.021 SEM, p = 0.049, t test; 0.05-1 mm ... 0.062 ± 0.024, p = 0.011), but not for the group of distant pairs (>1 mm ... 0 ± 0.15, p = 0.9)
Results, Noise Correlations in Area MSTd, paragraph 3; Figure S1How it will be judged
Pooled rnoise, groups by estimated distance as in the Experimental Procedures (single-electrode pairs set to 50 um, in the first group). One-sample t tests against zero (two-sided). Reproduced when the first two means are within tolerance and above zero at p < 0.05 and the third is not different from zero at p < 0.05; partially reproduced when two of the three groups match. Not testable if distances or the pairs over 1 mm are absent.
- methods-single-vs-multiple-electrodes
Noise correlations do not differ significantly between pairs recorded on one electrode (0.042 +- 0.02) and on different electrodes (0.033 +- 0.015), p > 0.7, t test.
noise correlations were slightly greater for pairs of neurons recorded from a single electrode (0.042 ± 0.02) than for pairs recorded from different electrodes (0.033 ± 0.015), this difference was modest and not significant (p > 0.7, t test)
Experimental Procedures, Electrophysiological Recordings, paragraph 2How it will be judged
A claim of absence, judged on a two-sample t test at p < 0.05 on pooled rnoise over all 179 pairs (the sample is not stated; all pairs is the reading, and the pairs under 1 mm a variant). Reproduced when the test finds no difference; the means are reported either way.
- fig2a-behavioral-learning
Heading discrimination thresholds of the five trained monkeys were high (over 10 degrees) early in training and decreased gradually to a plateau of 1 to 3 degrees.
These monkeys' heading discrimination thresholds (corresponding to 84% correct) were high ... at early stages of training, and gradually decreased to a plateau of only a few degrees
Results, Comparison of Noise Correlations in Trained and Naive Animals, paragraph 1; Figure 2AHow it will be judged
The range 1 to 3 degrees widened by 10% of each bound gives 0.9 to 3.3 degrees. Reproduced when each animal's threshold falls from above 10 degrees early in training to a plateau inside 0.9 to 3.3 degrees, the plateau read as the asymptote of an exponential fit to threshold against session, as in the legend. Needs the training histories, which the paper takes from earlier studies.
- fig3-mean-response-unchanged
The time course of the population average response to the preferred heading does not differ between trained (n = 146 neurons) and naive (n = 64) animals (p = 0.8, permutation test).
the time course of the populationaverage response to the preferred heading was indistinguishable between trained and naive animals (p = 0.8, permutation test
Results, Effects of Training on Tuning Curves, Variability, and Sensitivity of Single Neurons, paragraph 1; Figures 3A and 3CHow it will be judged
A claim of absence, judged on the paper's permutation test (Experimental Procedures, Permutation Test) at p < 0.05, on the mean firing rate at each cell's preferred heading in the 31 windows, for the neurons of the 127 pairs. Needs spike times. Whether the test was run per condition or on both together is open; the readings are a test per condition (visual and vestibular, as in Figures 3A and 3C) and one test on both concatenated. Reproduced when no reading's test finds a difference.
- fig3-fano-unchanged
Training has no significant effect on the time course of the Fano factor (p = 0.5, permutation test).
There was also no significant effect (p = 0.5, permutation test) of training on the time course of the Fano factor
Results, Effects of Training on Tuning Curves, Variability, and Sensitivity of Single Neurons, paragraph 1; Figures 3B and 3D; Figure S3How it will be judged
A claim of absence, judged on the paper's permutation test at p < 0.05. Fano factor per window and group as 10^intercept of a type II regression of log variance on log mean spike count with slope fixed at 1, values below 0.1 set to 0.1 (Experimental Procedures, Fano Factor; Figure S3). Which mean and variance enter the fit (per neuron and heading, or per neuron at the preferred heading) is open; per neuron and heading is the reading, as in the scatter plots of Figure S3. Per condition and on both conditions, as for fig3-mean-response-unchanged. Needs spike times. Reproduced when no reading's test finds a difference.
- fig4-tuning-width-visual
Median visual tuning width does not differ between naive (124.5 degrees) and trained (126 degrees) animals (p = 0.21, Wilcoxon rank-sum test).
There was no significant difference in median tuning width for the visual condition
Results, Effects of Training on Tuning Curves, Variability, and Sensitivity of Single Neurons, paragraph 2; Figure 4AHow it will be judged
A claim of absence, judged on a Wilcoxon rank-sum test at p < 0.05 on full width at half maximum, for neurons with significant heading tuning in the horizontal plane (one-way ANOVA p < 0.05, Results) in the single-electrode database (visual 466 naive and 526 trained, vestibular 216 naive and 340 trained, Figure S4 legend). How width is measured from the sampled tuning curve (interpolation) is not stated; linear interpolation of the tuning curve over the circle is the reading. Reproduced when the test finds no difference and both medians are within tolerance.
- fig4-tuning-width-vestibular
Median vestibular tuning width is slightly larger in trained (131 degrees) than in naive (121 degrees) animals (p = 0.045).
The difference in median tuning width was significant for the vestibular condition
Results, Effects of Training on Tuning Curves, Variability, and Sensitivity of Single Neurons, paragraph 2; Figure 4EHow it will be judged
Measure and selection as in fig4-tuning-width-visual. Reproduced when the trained median is larger with p < 0.05 and both medians are within tolerance.
- fig4-amplitude-unchanged
Training has no significant effect on tuning curve amplitude, visual (naive 35.4 against trained 31.8 spikes/s, p = 0.24) or vestibular (17.4 against 17.2 spikes/s, p = 0.36).
training did not have any significant effect on the distribution of tuning curve amplitudes in either the visual condition (naive ... 35.4 spks/s versus trained ... 31.8 spks/s, p = 0.24, Wilcoxon rank-sum test) or the vestibular condition (naive ... 17.4 spks/s versus trained ... 17.2 spks/s, p = 0.36)
Results, Effects of Training on Tuning Curves, Variability, and Sensitivity of Single Neurons, paragraph 2; Figures 4B and 4FHow it will be judged
A claim of absence, judged on Wilcoxon rank-sum tests at p < 0.05 per condition, on amplitude as peak minus trough of the mean tuning curve (Figure 4 legend), same neurons as fig4-tuning-width-visual. The values are read as medians, as for tuning width; means are shown as a variant. Reproduced when neither test finds a difference.
- fig4dh-lateral-more-sensitive
Neurons preferring lateral headings are more sensitive to heading variation around straight ahead than neurons preferring fore-aft headings (p << 0.001, factorial ANOVA).
lateral neurons were significantly more sensitive than fore-aft neurons for heading discrimination around straight ahead (p << 0.001, Factorial ANOVA, Figures 4D and 4H)
Results, Effects of Training on Tuning Curves, Variability, and Sensitivity of Single Neurons, paragraph 3; Figures 4C, 4D, 4G and 4HHow it will be judged
Neuronal threshold per neuron from Fisher information on the interpolated tuning curve at straight ahead (equations 2 and 3 for one neuron), with variance equal to the mean as equation 4 gives for i = j; how tuning curves are interpolated is not stated (readings, cubic spline and linear interpolation). Fore-aft is a preference within 45 degrees of 0 or 180, lateral within 45 degrees of -90 or 90 (Results). How the preferred heading is computed is not stated in this paper; the readings are the vector sum of the tuning curve and the peak of the interpolated curve. A factorial ANOVA on threshold with preference group and training group; whether it was run per condition (Figures 4D and 4H) or once with condition as a third factor is not stated, and both are run; whether thresholds were log transformed is open (raw and log are both run). Reproduced when lateral neurons have lower thresholds with p < 0.001 under one reading (in both conditions for the per-condition reading).
- fig4dh-training-no-sensitivity-change
Neuronal sensitivity does not differ between naive and trained animals for either fore-aft or lateral neurons (p > 0.5, factorial ANOVA), with no interaction (p > 0.3).
neuronal sensitivity was not significantly different between naive and trained animals (p > 0.5, factorial ANOVA) for either group of neurons, with no significant interaction effect (p > 0.3)
Results, Effects of Training on Tuning Curves, Variability, and Sensitivity of Single Neurons, paragraph 3; Figures 4D and 4HHow it will be judged
A claim of absence, judged on the factorial ANOVA of fig4dh-lateral-more-sensitive at p < 0.05 for the training effect and the interaction, per condition. Reproduced when neither is significant in either condition; the group means are reported either way.
- figs4-population-sensitivity-any-reference
Population sensitivity computed from the single-electrode database does not differ between naive and trained animals at any reference heading; the bootstrap 95% confidence bands largely overlap.
Confidence intervals are largely overlapping between naïve (blue) and trained (pink) animals
Results, Effects of Training on Tuning Curves, Variability, and Sensitivity of Single Neurons, paragraph 3; Figure S4How it will be judged
As in the Figure S4 legend, population threshold from equation 2 with independent Poisson neurons, 95% bands by bootstrap, per condition and group (visual 466 naive, 526 trained; vestibular 216 naive, 340 trained). The legend's "100 neurons were resampled with replacement" allows two readings, both run; populations of 100 neurons drawn with replacement (1,000 iterations, a standard choice), and the whole group resampled with replacement 100 times. The reference headings sampled are not stated; every 22.5 degrees is a standard choice. Reproduced when the naive and trained bands overlap at every reference heading in both conditions; partially reproduced when they overlap at most.
- fig5-noise-signal-positive
Noise correlation increases with signal correlation in both conditions (vestibular p = 0.0001, visual p = 0.0004, ANCOVA).
There was a significant positive correlation between rnoise and rsignal in both stimulus conditions (vestibular ... p = 0.0001; visual ... p = 0.0004, ANCOVA)
Results, Training Effects on the Noise-Signal Correlation Structure, paragraph 1; Figures 5A and 5BHow it will be judged
Per condition, an ANCOVA of rnoise on rsignal (continuous) and training group (categorical) with their interaction, over the 127 pairs. Which rnoise enters is open; the readings are the rnoise pooled across conditions (the measure the Results adopt before Figure 5) and the rnoise of the same condition as rsignal. Reproduced when, under one reading, the rsignal slope is positive with p < 0.05 in both conditions.
- fig5-slope-unchanged
Training does not change the slope of the relation between noise and signal correlation (vestibular p = 0.9, visual p = 0.9, ANCOVA interaction).
the slope of the relationship between rnoise and rsignal (Figures 5A and 5B) was not significantly affected by training (vestibular ... p = 0.9; visual ... p = 0.9, ANCOVA interaction effect)
Results, Training Effects on the Noise-Signal Correlation Structure, paragraph 2; Figure 5CHow it will be judged
A claim of absence, judged on the interaction term of the ANCOVA of fig5-noise-signal-positive at p < 0.05 in each condition, with the paper's pooled rnoise as the measure; the same-condition rnoise is shown as a variant. Reproduced when neither interaction is significant; the slopes and their 95% confidence intervals per group are reported either way.
- fig5-intercept-lower-trained
Training lowers noise correlation uniformly across signal correlations, a main effect of training (vestibular p = 0.02, visual p = 0.008, ANCOVA) with non-overlapping 95% confidence intervals of the intercepts.
training had a significant main effect on rnoise (vestibular ... p = 0.02; visual; p = 0.008 ANCOVA), and the 95% confidence intervals around the regression intercepts were non-overlapping for naive and trained animals
Results, Training Effects on the Noise-Signal Correlation Structure, paragraph 2; Figure 5DHow it will be judged
The training main effect of the ANCOVA of fig5-noise-signal-positive, and the 95% confidence intervals of the per-group regression intercepts. Readings for rnoise as in that claim. Reproduced when, under one reading, the naive intercept is higher with p < 0.05 in both conditions and the intervals do not overlap; partially reproduced when the main effect holds but the intervals overlap.
- fig5-regression-coefficients
The plane fitted to noise and the two signal correlations is rnoise = 0.12 rsignal,vestibular + 0.091 rsignal,visual + 0.072 in naive animals, with the same slopes and an intercept of 0.005 in trained animals.
How it will be judged
A plane fitted by type II regression (minimizing orthogonal distance, as in the Experimental Procedures) of pooled rnoise on vestibular and visual rsignal. How slopes come to be common to both groups is not stated; the readings are (a) one orthogonal fit to all 127 pairs with a separate intercept per group (rnoise centred per group before fitting the slopes) and (b) the fit to all pairs for the slopes, with each group's intercept as its mean of rnoise minus the slope terms, and (c) the slopes of an orthogonal fit to the naive pairs, used for both groups as the Results describe structure (1), with the trained intercept fitted with those slopes fixed. Ordinary least squares is shown as a variant. The tolerance of 10% on the trained intercept of 0.005 is tighter than any estimate can be; it is kept as the protocol default, and the 95% interval of the fitted intercept is reported beside it. Reproduced when, under one reading, all four values are within tolerance; partially reproduced when the slopes match and the naive intercept exceeds the trained one.
- fig6a-population-thresholds
In simulated populations, correlated noise as measured in naive animals raises the population threshold about 5-fold over independent noise at 2000 neurons, while the uniform reduction in trained animals has little effect on the threshold compared with naive.
For a simulated population of 2000 neurons, the predicted heading discrimination threshold was ... larger compared with the case of independent noise. Surprisingly, the uniform reduction in rnoise that we observed in trained animals (Figure 5) had little effect on predicted discrimination thresholds, as compared with naive animals
Results, Effect of Training on Population Coding Efficiency, paragraphs 1 and 2; Figure 6AHow it will be judged
Simulation from equations 2 to 5 with wrapped-Gaussian tuning (bandwidth 135 degrees), preferences uniform on the circle, congruent visual and vestibular tuning (so the two signal correlations are equal), reference heading 0, and the paper's coefficients (the coefficients fitted in fig5-regression-coefficients shown as a variant). Not stated are the tuning amplitude and baseline, the counting window, the headings over which rsignal is computed, and how bandwidth enters the wrapped Gaussian. The readings are amplitude 30 spikes/s on baseline 10 spikes/s over 1 s (a standard choice, not from the paper; Figure 4 reports median amplitudes of 17 to 35 spikes/s and no baseline), rsignal over the 8 headings 45 degrees apart and over a continuous circle, and bandwidth as full width at half height. The factor of about 5 is read from the extracted text, whose "approximately" sign is garbled. Reproduced when, under one reading, the ratio is within tolerance and the trained threshold at 2000 neurons is within 10% of the naive one; partially reproduced when correlated noise raises the naive threshold well above independent and trained and naive are within 10%, with the ratio outside tolerance.
- fig6b-slope-not-intercept
Population thresholds (256 neurons) are very sensitive to the slope of the relation between noise and signal correlation and only weakly sensitive to its intercept; a uniform increase in rnoise lowers the threshold slightly.
predicted thresholds were very sensitive to changes in the slope of the relationship between rnoise and rsignal. In contrast, changes in the intercept of the rnoise versus rsignal relationship had weak effects on predicted thresholds
Results, Effect of Training on Population Coding Efficiency, paragraph 2; Figure 6BHow it will be judged
The simulation of fig6a with 256 neurons, over a grid of slopes and intercepts for which rnoise stays within -1 and 1. The grid range is shown only in Figure 6B, whose axes could not be read from the text; slope 0 to 0.4 and intercept 0 to 0.2, which contain the measured naive and trained parameters, are a choice, not from the paper. Reproduced when, under the fig6a readings, the threshold changes more over the slope range than over an intercept range of the same width, and threshold decreases as the intercept rises at the measured slopes.
- fig7a-pair-sign-of-effect
For a single pair of neurons, reducing noise correlation lowers the discrimination threshold when signal correlation is positive and raises it when signal correlation is negative.
reducing the noise correlation between this pair of neurons results in a lower population threshold (red curve below blue curve) when the pair of neurons has positive rsignal. In contrast, reducing noise correlation increases the predicted threshold for negative rsignal
Results, Effect of Training on Population Coding Efficiency, paragraph 3; Figure 7A; Figure S7AHow it will be judged
Two-neuron simulation of equations 2 to 5 with the naive and trained coefficients, one neuron preferring -90 degrees and the other's preference swept from -90 to 90, tuning as in fig6a. Reproduced when the trained threshold is below the naive one for every pair with positive rsignal and above it for every pair with negative rsignal.
- fig7b-msdt-pairs
Decoding real MSTd pairs, the ratio of thresholds with rnoise = 0.1 and rnoise = 0 rises above 1 for pairs with positive signal correlation and falls below 1 for negative signal correlation (R = 0.49, p << 0.001, Spearman).
pairs of neurons with positive rsignal yield discrimination thresholds that increase with rnoise, whereas pairs with negative rsignal have discrimination thresholds that decrease with rnoise (R = 0.49, p << 0.001, Spearman rank correlation)
Results, Effect of Training on Population Coding Efficiency, paragraph 3; Figure 7BHow it will be judged
Needs each pair's tuning curves. Per pair, threshold from equations 2 to 4 at straight ahead with r = 0.1 and r = 0, from the pair's interpolated tuning curves; the ratio correlated with rsignal by Spearman rank correlation. Open are which condition (readings, visual and vestibular separately, and both pooled as two points per pair), which pairs (the 127 under 1 mm, or all 179), and the interpolation (cubic spline, linear). Reproduced when, under one reading, R is within tolerance and p < 0.001; partially reproduced when R > 0 with p < 0.05.
- fig7c-narrow-preferences
In a 256-neuron population, narrowing heading preferences to give more pairs with positive signal correlation (49%, 75%, 94%) makes the weaker noise correlations of trained animals improve coding, but at best the trained threshold is about 8% lower than the naive.
At best, however, the predicted population discrimination threshold for trained animals is ... lower than for naive animals
Results, Effect of Training on Population Coding Efficiency, paragraph 4; Figure 7C; Figure S7BHow it will be judged
The simulation of fig6a with 256 neurons and preferences uniform in [-180, 180], [0, 180] and [30, 150] degrees (Figure 7C legend), and narrower ranges around 90 as the figure's axis runs. The value of 8% and the percentages are read from the extracted text. Reproduced when, under one fig6a reading, the trained-over-naive improvement grows as the range narrows, its maximum is within tolerance, and the three proportions are within tolerance.
- fig8ab-choice-probability-lateral
Neurons preferring fore-aft headings have smaller choice probabilities than neurons preferring lateral headings (p = 0.019, t test), and choice probability does not differ between neurons preferring leftward and rightward headings (p = 0.11).
neurons with a preference for fore-aft movement show significantly smaller choice probabilities (p = 0.019, t test, Figures 8A and 8B) ... there was no significant difference in average choice probability between neurons preferring leftward and rightward headings (p = 0.11, t test)
Results, Possible rsignal Distributions in Area MSTd, paragraphs 2 and 3; Figures 8A and 8BHow it will be judged
Choice probabilities from the single-electrode database (n = 311, vestibular and visual pooled), groups by preferred heading as in Figure 4 (fore-aft within 45 degrees of 0 or 180, lateral within 45 degrees of -90 or 90; the preferred heading by the readings of fig4dh-lateral-more-sensitive). Two-sample t tests. Reproduced when fore-aft choice probabilities are smaller with p < 0.05 and the leftward and rightward means do not differ at p < 0.05.
- fig8cd-noise-not-preference
Noise correlation does not depend significantly on heading preference (p = 0.2, t test); lateral neurons have slightly smaller noise correlations than fore-aft neurons.
there was no significant ... dependence of noise correlation on the heading preferences of MSTd neurons (p = 0.2, t test)
Results, Possible rsignal Distributions in Area MSTd, paragraph 3; Figures 8C and 8DHow it will be judged
A claim of absence, judged on a t test at p < 0.05 of the noise correlation of lateral against fore-aft neurons, each pair entered twice, at each neuron's preferred heading (Figure 8C legend), groups as in Figure 4. The Results speak of trained animals, but the legend's 328 fixation points (164 pairs entered twice) exceed the 89 trained pairs under 1 mm (178 points), so the sample is not settled. The readings are the trained pairs under 1 mm, all trained pairs, and all 179 pairs, each with visual and with vestibular preference, with and without the 55 discrimination-task points. The test is run under each; the outcome is judged under the reading whose point count comes closest to 328 and 55, and the others are shown. Reproduced when that test finds no difference; the group means are reported either way.
- figs5-congruency
Noise correlation differs between naive and trained animals (p = 0.008) but is not affected by the congruency of the two units (p > 0.5), with no interaction (p > 0.5), two-way ANOVA; an ANCOVA finds no effect of congruency on the relation to signal correlation.
While the average noise correlation was significantly different between 'naïve' (blue bars) and 'trained' (red bars) animals (p=0.008, main effect of training, two-way ANOVA), it was not significantly affected by the congruency of the two units in each pair (p>0.5, main effect of congruency, two-way ANOVA)
Results, Training Effects on the Noise-Signal Correlation Structure, paragraph 3; Figure S5How it will be judged
Congruency index per neuron as the Pearson correlation of its visual and vestibular tuning curves; pairs CC, OO or CO by the signs (Figure S5 legend). Two-way ANOVA of pooled rnoise on training group and pair type, over the 127 pairs. Then, as in the legend, an ANCOVA of rnoise with rsignal as covariate and pair type and training group as factors; which rsignal enters (visual, vestibular) is not stated, and both are run. Needs tuning curves. Reproduced when the ANOVA training effect has p < 0.05 and none of the ANOVA congruency effect, the ANOVA interaction and the three ANCOVA congruency terms has p < 0.05 (claims of absence, judged on these tests).
- figs6-task-independent
For 41 pairs recorded in both tasks, noise correlations during fixation and during heading discrimination are correlated (R = 0.47, p = 0.002) and their means do not differ (p = 0.26, paired t test).
Noise correlations were significantly correlated across task conditions (R=0.47, p=0.002, Spearman rank correlation) and the mean values were not significantly different (p=0.26, paired t-test, n=41)
Results, Computation of Covariance Matrix, paragraph 2; Figure S6How it will be judged
Needs rnoise from the discrimination task, computed alike (the headings of that task differ; z-scoring per heading as in the Methods). The Results give p = 0.3 for the same comparison. The difference of means is a claim of absence, judged on a paired t test at p < 0.05. Reproduced when R is within tolerance with p < 0.05 and the paired test finds no difference.
- figs8c-renormalization
Renormalizing responses in blocks of 20 trials has no significant effect on noise correlations (R = 0.9 before against after), and the effect does not differ between naive and trained animals (interaction p = 0.7, group p = 0.9, ANCOVA).
This additional normalization had no significant effect on rnoise (p > 0.3, paired t test; R = 0.9, p < < 0.001, Spearman rank correlation, n = 127, Figure S8). More importantly, the effect of renormalization on noise correlations was similar in naive and trained animals (p = 0.7, interaction effect, p = 0.9, group effect, ANCOVA, Figure S8)
Experimental Procedures, Noise and Signal Correlations, paragraph 2; Figure S8CHow it will be judged
Needs trial-ordered responses. Renormalization of the z-scores in consecutive blocks of 20 trials (Zohary et al. 1994b); whether blocks run over all trials of the session or within each condition is open, and both are run. The Figure S8C legend gives p = 0.5 for the paired test. The ANCOVA is read as rnoise after renormalization on rnoise before (covariate) and training group. The absence parts are judged on the paired t test and the ANCOVA terms at p < 0.05. Reproduced when R is within tolerance and none of the three tests finds an effect.
- figs8ab-eye-movements
Naive animals fixate as accurately as trained animals and make fewer microsaccades (p << 0.001, t test).
we found that naive animals fixate as accurately as trained animals (Figure S8A). Indeed, naive monkeys as a group made significantly fewer microsaccades than trained animals (Figure S8B)
Discussion, Possible Explanations for the Effect of Training on Correlated Noise, paragraph 2; Figures S8A and S8BHow it will be judged
Needs eye position records. Deviation from initial fixation position, averaged over trials per animal in the middle 1 s; microsaccades as eye velocity peaks above 10 deg/s after smoothing with a 5 ms Gaussian (Figure S8 legend). The unit of the t test (trials or time bins per animal) is not stated; the readings are per trial and per animal. Reproduced when naive animals have fewer microsaccades with p < 0.05 under one reading and the mean deviation of naive animals lies within the range of the trained animals.
- discussion-stimulus-reduces-correlation
Noise correlations decrease in the presence of the stimulus compared with prestimulus baseline activity.
we observed that noise correlations decrease in the presence of a stimulus as compared with prestimulus baseline activity
Discussion, Properties of Noise Correlations in MSTd, paragraph 2How it will be judged
The paper shows no figure or test. Needs spike times. rnoise as for the main measure in a prestimulus window against the middle 1 s of the stimulus; the prestimulus window is not stated, and the reading is the fixation period before motion onset (100 to 200 ms, Experimental Procedures), with 500 ms before onset as a variant. The paper names no test; a paired t test over pairs is a standard choice, not the paper's. Reproduced when the mean prestimulus rnoise exceeds the stimulus rnoise by that test at p < 0.05.
Pipeline
- Paper
- waiting (not checked)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none