Scaling signatures of near-critical dynamics in large-scale human resting-state electrophysiology
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.
Ines: We're looking at the paper: "Scaling signatures of near-critical dynamics in large-scale human resting-state electrophysiology".
Marcus: Here's the gist of the paper. Scaling and tuning to criticality in resting-state human magnetoencephalography analyzes large-scale electrophysiological recordings of human resting-state brain activity and applies a renormalization group approach to track collective neural dynamics across spatial scales, finding that multiple observables exhibit robust scale-invariant behavior under coarse-graining Pruning redundant reasoning tokens From 1/f noise to neuronal avalanches, evidence of scaling in brain activity has been increasingly linked to tuning to or near criticality The concept of scaling is intimately related to the renormalization group (RG), in essence providing coarse-grained, simplified descriptions that generalize to classes of diverse physical systems Here, we analyze large-scale electrophysiological recordings (MEG) of human resting-state brain activity and apply a RG-inspired coarse-graining approach to track collective neural dynamics across spatial scales We find that multiple observables exhibit robust scale-invariant behavior under coarse-graining: activity variance and correlations grow according to power laws, covariance eigenspectra follow a characteristic scaling relation, and neuronal avalanche statistics remain invariant Using an analytically tractable neural network model, we show that the observed scaling signatures arise when the system operates slightly below criticality, and that the scaling exponents depend on the excitation–inhibition balance These findings demonstrate that RG-inspired scaling analysis can uncover signatures of critical dynamics in non-invasive human electrophysiology and suggest a principled route toward estimating excitation–inhibition balance from large-scale brain recordings How the training signal is built The coarse-graining procedure involves iteratively summing the activity of most correlated pairs of MEG variables until only a single variable remains For pedagogical purposes, we start with N = 4 MEG signals from sensors in the frontal and parietal areas (F46, F56, P44 and P45, Fig.
Ines: 1, left) The first step of the procedure is labeled k = 0 (original, non coarse-grained signals) Then, the sequence of events from each sensor is summed with the sequence of events of its most correlated sensor, leading to Nk = N/K = N/2 k new variables in step k = 1 (Fig.
Marcus: 1, middle) The procedure continues until Nk = 1, i.e., when only a single variable is left (Fig.
Ines: 1, right) At each step k of the procedure, we calculate the correlation matrix Ckij among the Nk variables and uniquely couple them according to its entries, as illustrated in Fig.
Title and authors: Ines: So we’re looking at this paper called "Scaling signatures of near-critical dynamics in large-scale human resting-state electrophysiology." It seems they're using a renormalization group approach to see if the brain activity fits a critical point.
Marcus: Right, it sounds like they are taking these big recordings from humans and trying to simplify them down by coarse-graining. The idea is that this simplification should reveal some underlying scaling behavior across different spatial scales in the brain's activity.
Yuki: From a population genetics perspective, when we see scaling in complex systems like neural networks, it hints at self-similarity across different levels of organization, which is something we often see in how species evolve. But this paper is looking specifically at human brains and their resting states.
Ines: The core finding is that they found multiple observables show robust scale-invariant behavior when you coarse-grain the data. They're tracking things like activity variance and correlations that grow according to power laws, and the covariance eigenspectra follow a specific scaling relation.
Marcus: That’s pretty concrete. They’re not just saying there’s some correlation; they're showing that these mathematical relationships hold up even when you simplify the data by summing up correlated pairs of signals iteratively until you get one variable left over.
Yuki: And it connects to something we hear about in neuroscience—the link between one/f noise and neuronal avalanches, which suggests tuning near criticality is important for brain dynamics <ref:2602.17820#pg1>. This paper is testing that idea on real human MEG data.
Ines: The analysis uses an analytically tractable neural network model to show that these scaling signatures only appear when the system operates slightly below criticality, and those specific scaling exponents depend on the excitation-inhibition balance of the network.
Marcus: That dependence on excitation-inhibition balance is what I find really interesting for a data scientist. It means if we want to measure this critical state, we need to know how that balance is set up in the system.
Yuki: So, it’s not just about finding a scaling law; it’s about using that scaling law—the exponents beta and alpha —as a tool to estimate the excitation-inhibition balance. That moves us from just observing dynamics to actually quantifying the network's state.
Ines: Exactly. The paper shows that their inferred adaptive Ising model, which mimics inhibitory neural inputs as negative feedback, fits the scaling in silence probability, activity variance, and correlation time with quantitative agreement for the exponents beta, alpha, and z measured in resting MEG data.
Marcus: That quantitative agreement is what makes it strong. It means the simplified model they built—this Ising model—is actually doing a good job of capturing the physics of the actual brain activity when you coarse-grain it.
Yuki: But we need to remember that they are using four MEG signals from sensors in frontal and parietal areas, like F46, F56, P44, and P45—that’s a specific set of data points. The question is how much of the human brain's resting state dynamics can we generalize from just those specific sensor locations?
Title and authors: Ines: That’s a fair point. And it brings up the methodology they use for coarse-graining, which starts with these four signals and iteratively couples them by summing the activity of most correlated pairs until you have just one variable left.
Marcus: So, the procedure itself is quite specific. They start with N signals, say four in this case. Then in step one, they take pairs—like F46 and F56—and sum their activity to get a new set of variables. It continues until you reach just one variable for the whole system.
Yuki: And they show that the distribution of activity rapidly approaches a non-Gaussian, exponential-like fixed form as the cluster size K increases, and that the probability of silence P zero(K) decays as a stretched exponential with beta equal to zero point eight two plus or minus zero point zero two <ref:2602.17820#pg3>.
Ines: That stretched exponential decay for silence probability is a key piece of evidence. It’s not just some random decay; it has this specific power-law shape that points towards the critical dynamics they are looking for in the phenomenological renormalization group analysis of MEG data.
Marcus: And they also found that the variance of this non-normalized cluster activity scales as a power law with K, specifically var(K) is proportional to K raised to the power alpha, which is one point three two plus or minus zero point zero three when looking at a hundred subjects.
Yuki: That scaling exponent alpha being close to one point three two is interesting because it aligns with previous observations in populations of neurons, suggesting a self-similar structure in these correlations that you’d expect near criticality.
Ines: They also looked at neuronal avalanche statistics and showed that they are invariant under the coarse-graining procedure, meaning the dynamics of these avalanches don't change much even when you apply more coarse-graining steps.
Marcus: That invariance is a strong statement about the robustness of the critical point they found. It suggests that whether you look at small clusters or large ones, the underlying dynamics are fundamentally similar in terms of how they behave.
Yuki: So, if we can confirm this invariance across different scales and observables—the activity distribution, silence probability, and avalanche statistics—it really strengthens the argument that resting-state brain activity is near a fixed point of a renormalization group.
Ines: The paper suggests that the relevant information about underlying neural dynamics is carried by high amplitude fluctuations within the Gaussian core of the signal amplitude distribution; smaller fluctuations are mostly noise.
Marcus: That helps explain why focusing on these scaling behaviors, rather than just raw signal magnitude, is important for capturing what’s actually driving the dynamics. It cuts through some of that noisy background activity we deal with in large cohorts.
Yuki: What this implies for us as a field is that instead of trying to identify one perfect critical point, we might be looking at a whole family of systems near criticality, and these scaling exponents help us map out where each system falls.
Title and authors: Ines: And the ultimate implication they are pushing is that alterations in the excitation-inhibition balance can be assessed by measuring those scaling exponents beta, alpha, and mu. That would be much more robust than trying to measure the power spectrum scaling, which you mentioned earlier might have crossover regimes.
Marcus: That's a huge practical implication for anyone working with large scale recordings. If you can use these specific exponents from the MEG data to tell how close the system is to that critical point based on its internal balance, that’s a much more direct measure than relying on spectral analysis alone.
Yuki: It connects back to the bigger picture of understanding neural populations across different scales, showing that even complex human brain activity can exhibit these universal scaling signatures when viewed through the lens of renormalization group theory.
Ines: So, to wrap up this paper, "Scaling signatures of near-critical dynamics in large-scale human resting-state electrophysiology," it shows that MEG resting-state dynamics is at a non-Gaussian fixed point of the phenomenological renormalization group.
Marcus: And because they showed that the inferred adaptive Ising model depends on the excitation-inhibition balance, we get a principled route toward estimating this key quantity from large scale brain recordings.
Yuki: I think what this means for future work is that we should look at how these scaling behaviors manifest in different brain regions or even under different conditions, because the paper shows these features are robust and scale-invariant across those analyses.
Ines: Yeah, and they suggest that alterations of the excitation-inhibition balance could be assessed by measuring those scaling exponents beta, alpha, and mu. That would be generally easier and more robust than estimating the scaling exponent of power spectra, where the scaling regime may not be unambiguously identifiable and crossover between different power-law regimes may exist.
Marcus: I agree. It’s about moving away from noisy spectral estimates toward these specific exponents that are tied to the system's internal structure, which seems much more stable across different types of data acquisition or even slightly different experimental setups.
Yuki: So, the main thing is that we should be looking for these scaling signatures in our own data and using them not just to describe what happened, but to quantify where the human brain dynamics are operating relative to criticality.
Ines: That's the focus then. The paper provides a framework for applying coarse-graining to non-invasive electrophysiology and suggesting a more principled way forward for studying critical dynamics in human brain recordings.
Marcus: It’s a solid piece of work that takes some very complex, noisy data and extracts meaningful, scale-invariant physics by employing the tools of the renormalization group.
Yuki: We’ll keep an eye on how this applies to other systems because if this holds for humans, it opens up a way to look for these scaling laws in other biological systems that exhibit similar complex dynamics.
The paper's summary: Marcus: Basically, they’re applying this coarse-graining technique to track how collective neural dynamics change across different spatial scales. They found that several things—like the variance of activity and the correlations—show robust scaling behavior even after you simplify the data down.
Ines: Robust scaling is key here. So, they aren't just seeing a random trend; they are seeing a mathematical structure that holds up when you zoom in or zoom out on the neural activity. It suggests there’s an underlying self-similarity in how the brain organizes itself across different sizes of clusters.
Yuki: And this links back to the idea of critical dynamics, which we’ve seen with things like neuronal avalanches, where we often see scaling behavior when systems are tuned near a phase transition point.
Ines: Exactly. The paper points out that they see these signatures in the activity variance growing as a power law with cluster size K, and the silence probability decaying in a stretched exponential way, which has this specific exponent beta of about zero point eight two.
Marcus: That exponent beta, when combined with the scaling of the variance—the alpha value which is around one point three two—that’s what they use to infer something deeper about how the network is balanced.
Yuki: And they connect that balance directly to the excitation-inhibition balance in a model called an adaptive Ising model. So, if you measure those scaling exponents, beta and alpha, you get a handle on how much inhibition versus excitation is pushing the system.
Ines: Right. It moves us past just observing noisy power spectra and gives us a way to quantify the network’s internal state through these scaling laws.
Marcus: It's powerful because they show that this model fits their real MEG data quite well, meaning it’s not just theory; it’s a good description of what’s happening in those large human cohorts.
Yuki: But we gotta remember, the authors use just four specific MEG signals from frontal and parietal areas to do this analysis. So, while the scaling is impressive for those sensors, we need to keep asking what that means for the whole brain dynamics.
Ines: That’s a fair caveat. And one of their findings is about neuronal avalanches—they showed those dynamics stay invariant even after they apply more coarse-graining steps, which really supports the idea that they are close to this critical fixed point.
Marcus: So, if you’re looking at complex data like this, you can use these scaling exponents to get a better handle on the system's physical state than just looking at the raw measurements.
Yuki: It means we can start using these scaling parameters as a way to characterize different states of brain activity rather than just one single measurement.
Ines: Exactly. And that leads us right into how these scaling exponents, beta, alpha, and mu, can be used to assess that excitation-inhibition balance in a much more robust way than estimating the power spectrum scaling.
The paper's improvements: Tom: So we're talking about how the authors suggest ways to make this study better after they’ve shown their main findings. They are proposing ways to get a more reliable picture of critical dynamics in brain recordings.
Ines: Right, they’re suggesting that instead of just looking at the scaling exponents themselves, we should use them as a way to measure the excitation-inhibition balance more directly.
Marcus: That makes sense. They flag that measuring those exponents beta and alpha is a much better way to quantify the E/I ratio than trying to get an exact power spectrum exponent, because the spectrum can have weird crossover points where it doesn't follow a single scaling law.
Yuki: From my perspective, that’s huge because it gives us a tool to characterize different brain states based on their internal wiring structure rather than just looking at raw signals which can be really noisy.
Ines: And they also suggest that the invariance of neuronal avalanches under coarse-graining is a key prediction. They think if we see that property hold true, it gives us strong evidence that the system is genuinely at a critical fixed point.
Marcus: So, they’re basically saying we should look for those avalanche dynamics staying consistent no matter how much we coarse-grain the data, which is a big check on their model.
Yuki: It’s about confirming that this self-similarity isn't just an artifact of the specific way they sampled or filtered the human recordings.
Ines: And they are pushing for future work to look at how these scaling behaviors hold up when you apply them to different brain regions, because we don't know if this is universal across all parts of the cortex.
Marcus: That’s a good point. The paper only used signals from frontal and parietal areas, so testing it on other parts would be a necessary next step to see if those scaling laws are general for the whole brain.
Yuki: And they also hint that this approach might be useful for studying other biological systems that show complex collective dynamics, because if it works here, we can look for these same signatures elsewhere.
Ines: So, the main improvement is a shift in focus: using scaling exponents to estimate network balance instead of just trying to fit a single power law to the spectrum.
Marcus: And that shifts the goal from describing what happened to quantifying *why* it happened by linking it back to the underlying network parameters.
Yuki: It’s about moving from observation toward a more principled method for studying critical dynamics in complex systems, whether that's human biology or something else entirely.
Conclusion: Ines: So to wrap up, this paper on "Scaling signatures of near-critical dynamics in large-scale human resting-state electrophysiology" shows that we can use a renormalization group approach to see if brain activity is operating at a non-Gaussian fixed point.
Marcus: And the main implication is that by measuring those specific scaling exponents, beta and alpha, we get a principled way to estimate the excitation-inhibition balance in the network.
Ines: It’s about moving past noisy power spectrum estimates and getting a more direct measure of how much inhibition versus excitation is driving the system.
Marcus: It’s powerful because they show that their inferred adaptive Ising model fits their real human MEG data pretty well, which gives us confidence in using this method for larger cohorts.
Yuki: What this means for population genetics is that we can use these scaling laws to characterize different states of neural activity based on the network structure itself, not just the raw signal measurements.
Ines: Exactly. They also confirmed that neuronal avalanche statistics are invariant under coarse-graining, which is a strong prediction suggesting the system is truly near criticality.
Marcus: So, if you’re looking at complex data like this, you can use these scaling exponents to get a better handle on the system's physical state than just looking at the raw measurements.
Yuki: It connects back to how we think about evolution; it suggests there are universal mathematical rules governing how these complex systems organize themselves across different scales.
Ines: Right. The paper shows that alterations in the excitation-inhibition balance can be assessed by measuring those scaling exponents, which is way more robust than estimating the power spectrum scaling where things often have crossover regimes.
Marcus: It’s about moving away from noisy spectral estimates toward these specific exponents that are tied to the system's internal structure, which seems much more stable across different types of data acquisition.
Yuki: I think this opens up a way for us to look for similar scaling laws in other biological systems that exhibit complex collective dynamics because if it works here, we can start testing it elsewhere.
Ines: It’s a solid piece of work on "Scaling signatures of near-critical dynamics in large-scale human resting-state electrophysiology," providing a framework for applying renormalization group theory to non-invasive electrophysiology.
Irem Topal, Anna Poggialini, Marco Dal Maschio, Daniele De Martino, Oren Shriki, Fabrizio Lombardi
Department of Biomedical Sciences, University of Padova · Biofisika Institute (CSIC,UPV-EHU) · Department of Industrial Engineering and Management, School of Brain Sciences and Cognition, Ben-Gurion University of the Negev
q-bio.NC, cond-mat.dis-nn, cond-mat.stat-mech, physics.bio-ph
Submitted: 2026-02-19
Updated: 2026-11-09
Code: https://github.com/lffrrnt/CoarseGraining
License: http://creativecommons.org/licenses/by-nc-sa/4.0/
Importance score: 83/100
The gist: Scaling and tuning to criticality in resting-state human magnetoencephalography analyzes large-scale electrophysiological recordings of human resting-state brain activity and applies a…
Key concepts
- Renormalization Group (RG)
- A mathematical tool used to simplify complex systems by describing their behavior across different spatial scales. In this study, it helps track collective neural dynamics by repeatedly summarizing large amounts of data into smaller, coarser descriptions to reveal underlying scale-invariant patterns.
- Coarse-graining
- The process of simplifying data by iteratively summing the activity of correlated pairs until only a single variable remains. This technique reveals how neural activity structures itself across different spatial scales, showing that key properties like variance and correlations follow predictable power laws.
- Criticality
- A state where a system exhibits scale invariance, meaning its behavior looks similar regardless of the scale at which it is observed. The analysis suggests resting-state brain activity is near this critical point, characterized by specific scaling exponents that depend on the excitation/inhibition balance.
Terminology
Summary
Scaling and tuning to criticality in resting-state human magnetoencephalography analyzes large-scale electrophysiological recordings of human resting-state brain activity and applies a renormalization group approach to track collective neural dynamics across spatial scales, finding that multiple observables exhibit robust scale-invariant behavior under coarse-graining
Pruning redundant reasoning tokens
From 1/f noise to neuronal avalanches, evidence of scaling in brain activity has been increasingly linked to tuning to or near criticality The concept of scaling is intimately related to the renormalization group (RG), in essence providing coarse-grained, simplified descriptions that generalize to classes of diverse physical systems Here, we analyze large-scale electrophysiological recordings (MEG) of human resting-state brain activity and apply a RG-inspired coarse-graining approach to track collective neural dynamics across spatial scales We find that multiple observables exhibit robust scale-invariant behavior under coarse-graining: activity variance and correlations grow according to power laws, covariance eigenspectra follow a characteristic scaling relation, and neuronal avalanche statistics remain invariant Using an analytically tractable neural network model, we show that the observed scaling signatures arise when the system operates slightly below criticality, and that the scaling exponents depend on the excitation–inhibition balance These findings demonstrate that RG-inspired scaling analysis can uncover signatures of critical dynamics in non-invasive human electrophysiology and suggest a principled route toward estimating excitation–inhibition balance from large-scale brain recordings
How the training signal is built
The coarse-graining procedure involves iteratively summing the activity of most correlated pairs of MEG variables until only a single variable remains For pedagogical purposes, we start with N = 4 MEG signals from sensors in the frontal and parietal areas (F46, F56, P44 and P45, Fig. 1, left) The first step of the procedure is labeled k = 0 (original, non coarse-grained signals) Then, the sequence of events from each sensor is summed with the sequence of events of its most correlated sensor, leading to Nk = N/K = N/2 k new variables in step k = 1 (Fig. p. 1 The initial step of the procedure is labeled k = 0 (original, non coarse-grained signals). Then,
the sequence of events from each sensor is summed with
the sequence of events of its most correlated sensor, leading to Nk = N/K = N/2
k new variables in step k = 1
(Fig. The new variables are then normalized in such a way that the average number of events in
non-empty time bins is one, as in the initial step k = 0
(Appendix B). The procedure continues until Nk = 1,
i. The procedure continues until Nk = 1,
i. , when only a single variable is left (Fig. Hence, the variables in step k are clusters
of size K and are labeled as x
(K)
i
(i = 1, . To implement this procedure on MEG recordings, at
each step k, we calculate the correlation matrix C
k
ij
among the Nk variables and uniquely couple them according to its entries, as illustrated in Fig.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the gains come from
The analysis reveals several key scaling behaviors that are robust under coarse-graining Specifically, we find that the distribution of activity rapidly approaches a non-Gaussian, exponential-like fixed form for increasing values of the cluster size K (Fig. 2A) The probability of silence P0(K) decays as a stretched exponential, P0(K) ∝ exp(−aKβ), with β = 0. p. 2 This
process is repeated iteratively until only one variable is left (right, k = 2). decays as a stretched exponential, P0(K) ∝ exp(−aKβ
),
with β = 0. decays as a stretched exponential, P0(K) ∝ exp(−aKβ
),
with β = 0. 02 (mean ± SD; Fig. 02 (mean ± SD; Fig. 2B, individual
subject; Fig. In fact, we observe that the variance
of the non-normalized cluster activity scales as a powerlaw of K, V ar(K) ∝ Kα, with α = 1. 03 (mean ±
SD; n = 100).Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Scaling in collective neural activity: from populations of neurons to fields
Evidence of a non-Gaussian fixed point in collective neural dynamics has recently been reported for population of neurons in the hippocampus [31], and over distinct regions of the mouse cortex [32, 33] Our analysis shows that these fields, although originating from filtered and highly non-linear superposition of neural/synaptic activity [36, 46], are collectively described by scaling laws that are remarkably similar—and in some cases comparable—to those observed in small populations of neurons We also found robust scaling in the eigenspectra of the covariance matrices, with an exponent µ that correlates with the scaling exponents β and α However, a close comparison with fMRI is difficult due, for instance, to the specific choice in the discretization threshold [34, 37]
Coarse-graining and avalanche criticality
The coarse-graining analysis suggests that resting-state brain activity is close to a fixed point of the dynamics characterized by scale invariance, i.e. criticality In Fig. p. 6 On the other hand, γ and
the avalanche duration exponent, αt, remained nearly
unchanged (Fig. Excitation/inhibition balance modulates scaling
exponents near criticality
To understand the emergence of scaling in restingstate brain activity in relation to neural dynamics
and criticality, we simulated resting-state activity using a recently proposed interpretable neural network
model [39]. Neuronal avalanches are labeled as
Ai and defined as sequences of time bins continuously
populated with at least one extreme event (blue dots)
on any of the MEG sensors or variables (Appendix C). As we proceed with coarse-graining, we notice that the
number of variables decreases, as well as the number
of extreme events (blue dots), whereas the time windows occupied by the avalanches (i. As we proceed with coarse-graining, we notice that the
number of variables decreases, as well as the number
of extreme events (blue dots), whereas the time windows occupied by the avalanches (i. the durations)
remain unchanged (Fig. the durations)
remain unchanged (Fig. To quantify the dynamics of avalanches across coarse-graining steps, we plot
the distributions of avalanche sizes and durations as a
function of K (Fig. To quantify the dynamics of avalanches across coarse-graining steps, we plot
the distributions of avalanche sizes and durations as a
function of K (Fig. We observe that the largest
avalanche size decreases as K increases (Fig.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
E/I balance and scaling behavior in the resting-state
The inferred adaptive Ising model, in which a negative feedback mimics the effect of inhibitory neural inputs [39], closely reproduced the scaling in silence probability, activity variance and correlation time, with the exponents β, α and z in quantitative agreement with the values measured in the resting MEG In its simplest configuration, the feedback represents an inhibitory drive that influences neurons on a certain time scale controlled by c (Appendix E) Absence of negative feedback mimics a scenario without inhibition, where neural dynamics is solely controlled by excitatory neuronal interactions (Jij)
Network dynamics and phenomenological renormalization
The inferred model closely reproduces resting-state activity under coarse-graining First, the distribution of activity, QK(x), tends to a non-Gaussian fixed form for increasing cluster size K (Fig. 5A) The probability of silence scales non-trivially with K, i. p. 7 First, the
distribution of activity, QK(x), tends to a non-Gaussian
fixed form for increasing cluster size K (Fig. 5A), and
the probability of silence scales non-trivially with K, i. This
process is repeated iteratively until only one variable is left (right, k = 2). decays as a stretched exponential, P0(K) ∝ exp(−aKβ
),
with β = 0. P0(K) = e
−aKβ
, with β ≈ 0. 8 in agreement with the
empirical value (Fig. 8 in agreement with the
empirical value (Fig. Similarly, and in line with the
relation between exponents found in empirical data, the
activity variance grows as a power-law with K, with an
exponent α ≈ 1. Similarly, and in line with the
relation between exponents found in empirical data, the
activity variance grows as a power-law with K, with an
exponent α ≈ 1. 3, as observed in real data (Fig.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Conclusion
Overall, the non-Gaussian form of the distributions of activity and the scaling behaviors observed in all analyzed quantities indicate that MEG resting-state dynamics is at a non-Gaussian fixed point of the phenomenological renormalization group The adaptive Ising model, with its two parameters inferred from data, suggests that empirical scaling exponents depend on the E/I balance of the network, paving the way for robust assessment of this key quantity in neuroscience research
The paper demonstrates that resting-state brain activity in humans is close to a non-trivial fixed point of a phenomenological renormalization group Activity in coarse-grained clusters approaches a fixed non-Gaussian distribution as their sizes increase, while the probability of silence P0 (no activity) decreases sub-exponentially The variance and autocorrelation of cluster activity both scale as a power-law of the cluster size, indicating a scale invariant structure of correlations in resting-state brain activity This was further confirmed by the size-independent scaling behaviors
The authors investigate how neuronal avalanche dynamics evolve under the coarse-graining procedure studied above, and show that they are invariant The analysis shows that the difference αt−1 τs−1 −γ tends to zero for increasing cluster sizes (Fig. S17), indicating that the system is getting closer to criticality
The inferred adaptive Ising model, with its two parameters inferred from data, suggests that empirical scaling exponents depend on the E/I balance of the network, paving the way for robust assessment of this key quantity in neuroscience research The authors note that alterations of the E/I balance could be assessed by measuring one of the scaling exponents β, α and µ
The results suggest that one needs to stay rather close to criticality to observe the non-trivial scaling behaviors of resting-state MEG The model suggests that feedback plays an important role in reproducing the scaling behaviors of resting-state brain activity in humans
The study concludes that robust scaling behaviors and data collapses in resting-state MEG result from the intrinsic structure of the underlying correlations Random coarse-graining might not fully destroy the scaling features of the original data because correlations among MEG sensors are non-negligible over large areas, i.e. they are non-local and decay rather slowly with distance [9]
The analysis shows that in the limit of continuous signals (variables), we observe that the distribution of activity deviates from a Gaussian and shows long tails (positive and negative) before coarse-graining and for a number of coarse-graining steps (Fig. S11)—nearly independently of K—and converges to its Gaussian core for large K
The authors conclude that the relevant information about underlying neural dynamics is carried by high amplitude fluctuations In contrast, small amplitude fluctuations within the Gaussian core of the signal amplitude distribution are mostly ‘noise’
The analysis suggests that for MEG data, the two procedures may be equivalent (at least to some extent) The relation between scaling in coarse-grained brain activity and typical measures of brain criticality (or nearcriticality) based on avalanche dynamics remain to be studied, and is not clear whether they provide a complementary picture of brain criticality
The authors conclude that the E/I ratio is also linked to the scaling behaviors of resting-state activity under coarse-graining This would be generally easier and more robust than estimating the scaling exponent of (often noisy) power spectra, where the scaling regime may not be unambiguously identifiable and crossover between different power-law regimes may exist [50]
The study suggests that alterations of the E/I balance could be assessed by measuring one of the scaling exponents β, α and µ This would be generally easier and more robust than estimating the scaling exponent of (often noisy) power spectra, where the scaling regime may not be unambiguously identifiable and crossover between different power-law regimes may exist [50]
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved analysis of brain activity in resting-state electrophysiological recordings by applying a renormalization group inspired coarse-graining approach to track collective neural dynamics across spatial scales. This allows for uncovering signatures of critical dynamics in non-invasive human electrophysiology and suggests a principled route toward estimating excitation inhibition balance from large scale brain recordings.
-
Enhanced ability to estimate the excitation inhibition balance by using scaling exponents from MEG data, specifically the relationship found between β and α, which can be used to infer the E/I ratio from large scale brain recordings. This is more robust than estimating power spectrum scaling exponents due to potential crossover phenomena.
-
Verification of neuronal avalanche dynamics invariance under coarse-graining for resting-state MEG signals by showing that
neuronal avalanches are invariant under the same coarse-graining procedure.
This provides a key prediction of the renormalization group for systems at criticality in non invasive human electrophysiology.
Abstract
From 1/f noise to neuronal avalanches, evidence of scaling in brain activity has been increasingly linked to tuning to or near criticality. The concept of scaling is intimately related to the renormalization group (RG), in essence providing coarse-grained, simplified descriptions that generalize to classes of diverse physical systems. Following the RG idea, scaling laws have been reported in populations of spiking neurons at microscopic scales. Whether similar scaling principles govern large-scale neural activity in the human brain and how they relate to underlying neural physiology remains unresolved. Here, we analyze large-scale electrophysiological recordings (MEG) of human resting-state brain activity and apply a RG-inspired coarse-graining approach to track collective neural dynamics across spatial scales. We find that multiple observables exhibit robust scale-invariant behavior under coarse-graining: activity variance and correlations grow according to power laws, covariance eigenspectra follow a characteristic scaling relation, and neuronal avalanche statistics and scaling relationships are conserved. Using an analytically tractable neural network model with parameters inferred from data, we show that the observed scaling signatures arise close to, but slightly below, criticality, and that the scaling exponents can be modulated by network inhibitory drives. These findings demonstrate that RG-inspired scaling analysis can uncover signatures of critical dynamics in non-invasive human electrophysiology and suggest a principled route toward estimating excitation-inhibition balance from large-scale brain recordings.
Sources
Related papers
- Self-Supervised Learning of Local Predictive Directions as a Neural Computation
- Synchronized Hearts, Desynchronized Brains: A Nonlinear Dynamics Perspective for Function and Failure
- Message interaction dynamics covary with brain volume across mammalian connectomes
- Planning as Dynamics Relaxation: Hippocampal Recurrent Network Realizes Optimal Goal-Directed Navigation
- The Platonic brain bridge hypothesis: brain networks as architectural priors for multimodal large language models
- BrainPFN: Amortized Brain-Behaviour Prediction via Prior-Data Fitted Networks