The 1st online meeting

[MakotoMiyakoshi]

When
Feb 13, 2026 11:00 AM Eastern Time (US and Canada)

Where
[ended]

Time table
11:00-11:10 Keynote: FOOOrier transformation: A shortcut to FOOOF's heart (Makoto)
11:10-11:20 Documented and undocumented 1/f contributors (Makoto)
11:20-11:30 Additive vs. Multiplicative [TBD] (Mate)
11:30-11:40 How to separate the periodic & aperiodic components (Cedric)
11:40-11:50 A novel median-variant approach [TBD] (Eugen)
11:50-12:00 Blind source separation and RNN [TBD] (Jamie)
12:00-12:10 Discussions and questions (Do we want to host a special issue? etc..)

Statistics

Recorded video file
https://ucsd.zoom.us/rec/share/isLCQKjTHqs5ZQzGXSAXPxbrNAFWFYbgsF6iMzAJ_qXSQvomJH9-2R_lbb0iQkif.64Z7ZXaCuwhYsmzO?from=hub
Passcode: =0VP&M+p

When you choose 'Download (4 files)' button on the top right, you can read all messages posted to the Chat window.

The video and chat text are shared for internal use only.
Redistribution or reposting in any form is strictly prohibited.

Summary by ChatGPT5.2

1) Keynote: “FOOOrier transformation: a shortcut to the heart of FOOOF” (Makoto Miyakoshi)

Core didactic move

• 10-Hz sinusoid in time domain → a peak at 10 Hz in frequency domain.
• Dirac delta (infinitely narrow and high spike) in time domain → flat spectrum in frequency domain.

Practical heuristic

• Narrower in time domain ⇒ flatter PSD in frequency domain (closer to Dirac delta)
• Broader in time domain ⇒ steeper PSD in frequency domain (further from Dirac delta)

---

2) “Documented and undocumented sources of 1/f: four contributors across spatial scales" (Makoto Miyakoshi)

Main claim: AMPA/GABA alone is insufficient. At least four mechanisms can generate 1/f-like behavior in scalp EEG/LFP.

2.1 Spatial averaging / cancellation

(Nunez & Srinivasan, Electric Fields of the Brain (EFB))

• Below beta: broader coherent sources → less cancellation → more power.
• Above beta: smaller, less synchronous sources → more cancellation → reduced power.
• Net effect: apparent 1/f-like roll-off in scalp EEG.

(In the video, Makoto calls a diameter as a radius by mistake)

2.2 Cable theory / dendritic filtering

(Rall et al., 1967)

• Synaptic input farther from soma ⇒ stronger low-pass filtering.
• Warning: this can mimic “E/I” interpretations if slope alone is considered.

2.3 AMPA vs GABA-A PSP time-course differences

(Gao et al., 2016)

• Faster (narrower) excitatory PSPs vs slower (broader) inhibitory PSPs.
• Leads to flatter vs steeper spectral profiles (as in Gao/Donoghue-type modeling logic).

2.4 Slow soma–dendrite ionic concentration / open current-loop dynamics

(Sætra et al., 2021; Halnes et al., 2024)

• Hypothesized very slow modulator (<1 Hz, sometimes into delta range).
• Linked to multi-second baseline shifts in a 40 Hz ASSR dataset (red vs blue groups, including Fragile X).

Meta-message

The “aperiodic” component likely mixes multiple spatial scales (micrometers → centimeters). Mechanistic mapping is therefore nontrivial (“Matryoshka doll” metaphor).

---

3) Additive vs multiplicative models (Mate Gyurkovics)

Core issue

Baseline correction via dB conversion implies division by baseline → multiplicative framing.

If the true relationship is additive (signal + noise), division can systematically deflate estimates when noise differs across conditions/subjects.

Extension to spectral decomposition

• Many methods appear subtractive in log-space.
• In linear space, due to log identities, they become divisive.
• Estimating oscillatory peak magnitude can therefore be biased by varying 1/f background, depending on the assumed model.

---

4) GLM-like single-trial separation (Cedric Cannard, presented by Makoto on his behalf)

Concept

Model single-trial spectra such that:

• Aperiodic/background term (intercept + slope)
• Localized oscillatory term (e.g., alpha peak)

are estimated separately.

Simulations

• Ground-truth oscillation held constant.
• Background varied.
• Some methods recovered true oscillation; others showed dependency on background.
• Only FOOOF appeared robust/uncorrelated in that simulation battery.
• IRASA unexpectedly failed (possibly due to simulation/setup issues).

---

5) “Maximally simple” robust slope estimator (Eugen Masherov)

Reconstructed idea:

• Avoid many hyperparameters (critique of FOOOF sensitivity and IRASA behavior).
• Estimate aperiodic exponent via many pairwise log-log slopes between frequency pairs.
• Aggregate slopes using robust central tendency (median or trimmed mean).
• Compare methods by embedding resulting parameter vectors into 2D via distance-preserving projection (e.g., MDS/t-SNE/UMAP-like approach).

Algorithmic reconstruction can be provided if needed.

---

6) RNN approach: ERP BSS → PSD (Jamie O'Reilly)

Approach

• RNN trained (supervised) to reproduce labels.
• For ERPs: outputs + final linear mixing layer interpreted as scalp maps and source time courses.

PSD attempt

• Model trained on PSD labels.
• Tended to smooth spectra and miss narrow peaks (e.g., 12/15 Hz SSVEP).

Hypothesis: spiky spectral features contribute little to MSE loss; model preferentially fits broad structure.

---

7) Discussion themes

Clinical relevance (Michael)

• 1/f is framed as foundational, not niche.

“Other receptors?” (Jamie)

• Response: mechanism is PSP shape, not receptor label per se (Antonio reinforced this).

Cross-disciplinary usage

• FOOOF-like peak/background separation used outside neuroscience (e.g., astronomy PSD analysis).

Biological validation (Gian Marco)

• Perturbation tests proposed:
  • Pharmacology
  • In vitro networks
  • Clinical syndromes
    to validate E/I interpretation.

Multi-mechanism view (Evie)

• 1/f likely reflects structural + functional contributions.
• Useful as a marker, not necessarily a single-cause phenomenon.

Frequency range concerns (Antonios)

• Current practice often fits ~1–45 Hz.
• Very slow processes (0.1–3 Hz) may be important.
• Emphasis on clarifying what EEG/LFP physically represents.

[EugenMasherov]

First of all, a huge thank you to organizer Makoto Miyakoshi!
Flocks of facts and ideas, scattered along the branches of the discussion tree, have converged and begun to sing in unison. I think this is the beginning of something magnificent and grandiose.
According to the reports.
I completely agree with the thesis that all four of the aforementioned mechanisms influence the formation of aperiodic activity. I believe we shouldn't argue, like Heraclitus and Thales, over whether water or fire lies at the root of the Universe. Instead, like Aristotle, we should say that there are four primary causes, plus a fifth, mysterious quintessence. However, as a fifth factor, and a very unpleasant one, I would consider the influence of artifacts, the complete elimination of which is extremely difficult, especially in the conditions of EEG recording in a clinic. This includes low-frequency activity, which may be associated with galvanic skin response, electrode polarization, or patient movement, and which increases the spectrum in the low-frequency region and thereby increases the slope. It also includes high-frequency activity, particularly muscle activity, whose spectrum is close to white noise and thus reduces the slope, since a small addition changes the spectrum little in the low-frequency region, but even a relatively large change in the high-frequency region. This may hinder the fundamentally correct idea of ​​calculating using a wider frequency band than a typical clinical EEG, since additional frequency bands are more heavily contaminated with artifacts. However, even if we observe the cumulative effect of several mechanisms, and one burdened by artifacts, comparing the indicator values ​​with clinical data can provide a very useful tool (but this is a task for a large-scale, possibly multicenter, project). The presentations on the principles of FOOOF, the differences between additive and multiplicative approaches, the comparison results of different algorithms, and automatic separation based on a neural network were extremely interesting and useful. Regarding my presentation, I apologize once again for my extremely poor English.
As for my remark that if you hold a hammer, everything looks like nails, while if you hold a Fourier transform, everything looks like sinusoids, this isn't a complaint, but an attempt to point out that the research tool fundamentally influences the result, even if we don't pay attention to this influence. This leads me to prefer a uniform tool, not the best one, over one that is optimally configured for each specific case.

[MakotoMiyakoshi]

Hi @EugenMasherov

I believe we shouldn't argue, like Heraclitus and Thales, over whether water or fire lies at the root of the Universe.

Somehow, I enjoy every piece of analogy you throw.

if you hold a Fourier transform, everything looks like sinusoids,

I heard a variation of this famous quote (from Adler, right?) from my mentor.

I understand that pure measurement of 1/f is like an idealized assumption. Whitham et al. (2007) shows a real horror for 1/f researchers.

In fact, I realized I forgot to count 'thalamocortical dysrhythmia' as the fifth element of 1/f-ness. This thing, known by Llinas' works, makes 1/f curve as if GABA_A-R became dominant.

[Ajarn-Jamie]

The results of RNN-BSS of PSD are starting to look a bit more promising. After running the model optimization for many more iterations to better fit the labels, several consistent and expected (e.g. 12 Hz vs. 15 Hz SSVEP responses) spectral sources can be identified, e.g.:

If the spectral features of EEG (or mechanisms of 1/f-ness) are topographically distinct, this method could provide a way to separate them.

@EugenMasherov the same thought has been recurring to me for some time. My hammer is RNN 😆

[EugenMasherov]

I understand that pure measurement of 1/f is like an idealized assumption. Whitham et al. (2007) shows a real horror for 1/f researchers.

Are you referring to this article?
Scalp electrical recording during paralysis: Quantitative evidence that
EEG frequencies above 20 Hz are contaminated by EMG.
As far as I understand, such observations led to the gamma rhythm, originally identified by EEG pioneers, being forgotten (and a gap appeared in the list of rhythms—alpha, beta,..., delta). It was only after recordings were obtained during surgery (both from the scalp with muscle relaxation and directly from the cortex) that the term was reintroduced. But I have some doubts about how dire the situation is. Cisatracurium, which the authors of this article used for muscle relaxation, does not penetrate the blood-brain barrier, but its metabolic product, laudanosine, does. I can't estimate its impact, but I'll quote from a reference book:

It has been shown to interact with GABA receptors, glycine receptors, opioid receptors, and nicotinic acetylcholine receptors, but not benzodiazepine or muscarinic receptors, which are also involved in epilepsy and other types of seizures.

I don't know whether it could reduce high-frequency activity. But I can't rule out such an explanation.
At one point, the late Professor Shchekutyev, graduate student Volinsky, and I (along with him) tried to devise a better way to eliminate muscle interference than filtering. The idea was that muscle artifact is more localized and uncorrelated in adjacent leads. It looked very appealing graphically, but we couldn't prove that the distortion was less than with filtering.
https://link.springer.com/article/10.1134/S0362119709040161
On the other hand, if one does not try to expand the range, limiting it to 35 Hz or similar, it is possible that even if the influence of the myogram is significant at high frequencies, the error in estimating the slope of the aperiodic component may be acceptable.

[EugenMasherov]

Trying to assess the effect of filters.
I took a real EEG recording and calculated the slope using different spectrum boundaries (which isn't the same as filtering, since real filters have a smoothly falling amplitude-frequency response rather than a sharp zero, but it does give an approximate picture).

Apparently, high-frequency filter settings, given that HF ​​filters vary over a narrower range than LF filters, have a very significant effect on slope estimates, while they have a less significant effect in the low-frequency range. From a practical standpoint, this means that, to ensure comparability of estimates, filter values ​​must be specified and, possibly, the filter type and order must be taken into account. This further raises the issue of the influence of muscle artifact.

[MarjanZamani1990]

Hi @MakotoMiyakoshi

First, I would like to sincerely thank you for your generosity and for providing us with this valuable opportunity.

Regarding the recent Zoom session, I must admit that I have been somewhat away from the discussions and have not yet studied the newer topics. Due to my limited English speaking skills and the time I have been absent, I preferred to listen and learn rather than actively participate.

If you are still willing, it would be a great honor for me to contribute to your paper.

In the context of periodic and aperiodic EEG components, could the following spectral analysis approach be used to investigate the contribution of each component to the overall signal and address the research questions below?

Proposed approach:

1. Compute the full EEG power spectral density (PSD)
2. Re-estimate the PSD after excluding a target frequency band
3. Isolate and analyze the PSD of only the target frequency band

By comparing these spectra, structural similarities, parameter shifts, and potential interactions between components could be quantified.

Research questions:

Does task-related activity overlap with specific frequency sub-bands and their corresponding brain regions?

Can the contribution of each brain region to the overall EEG signal be determined?

Additionally, by combining this spectral PSD approach with the RNN_BSS framework(@Ajarn-Jamie ), could we also investigate whether the relationship between periodic and aperiodic components is better characterized as additive or multiplicative at the generative level?

[EugenMasherov]

To answer this question definitively, we need to clarify what constitutes an alpha rhythm. The standard definition, in addition to a frequency of 8-13 Hz, includes a posterior region, a waveform close to sinusoidal, and, as a key feature, a response to eye opening. From this perspective, there is only one alpha rhythm, since all others do not respond to eye opening. However, what responds to eye opening does not always fall within the alpha rhythm frequency range. It can be slowed to 6-8 Hz, either as a mild pathology or as an age-related norm in children, and can be accelerated to 14-16 Hz. Perhaps it is worth distinguishing between the alpha-frequency rhythm (8-13 Hz) and the alpha-like rhythm that responds to eye opening, and considering the intersection of the two as the undisputed alpha rhythm. It seems to me (and I'm trying to make a similar point in the article with Alexandrov and Ivanov, which I sent you) that the corticothalamic mechanism operates in relation to the occipital alpha rhythm, but there are rhythms in this frequency band that have a different origin. Perhaps my opinion is too radical and mechanistic, but it seems to me that the most visually noticeable sinusoidal oscillations, which seem to be the main part of the EEG, are in fact associated not with the transmission of information, but with the regulation of the state of brain tissue. A narrowband signal carries a minimum of information, the maximum information is carried by white noise, and a 1/f spectrum can be optimal for transmitting information under certain constraints. At the same time, sinusoidal oscillations arise naturally as a response of an integral regulator. The oscillation frequency reflects the state of the regulated object, decreasing with deterioration (for example, during hypoxia) and increasing with an increase in oxygen supply, for example, during hyperbaric oxygenation. This is likely just one of many regulators operating simultaneously, some of which are associated with neurons, others with glia (and possibly also with blood flow regulation), so we see multiple oscillations superimposed on each other. However, under conditions of intense brain activity, the flood of disturbances disrupts the regularity of the oscillations, and we see the sinusoid, like the narrow peak in the spectrum, only in idle mode. As a speculative hypothesis, sleep spindles appear when blood flow and oxygenation correspond to wakefulness levels, but the impulse flow has decreased, so the sinusoids are of a fairly high frequency, while sleep delta activity occurs when blood flow has decreased and oxygen supply is reduced.

[EugenMasherov]

@MarjanZamani1990

Here are links to Mate's key works. https://www.sciencedirect.com/science/article/pii/S1053811921004699?via%3Dihub https://www.jneurosci.org/content/42/37/7144#libraryItemId=14213983

This article contains the following paragraph:
The small negative correlation between narrowband (alpha) and broadband (1/f) activity we did observe was therefore likely artifactual, and consistent with predictions made under the hypothesis of a true zero correlation between these two types of activities, but inconsistent with a positive correlation between these activities, predicted by the multiplicative model.
And perhaps it means that the "addition or multiplication" question is incorrect, that there is a third option, "substitution." That periodic activity in everything arises as idle, at least in a small region, displacing the working activity, which manifests itself as a broadband signal with a 1/f spectrum.
And instead of subtracting or dividing, we need to exclude fragments with periodic activity, processing them separately.

[MakotoMiyakoshi]

@EugenMasherov,

Perhaps my opinion is too radical and mechanistic, but it seems to me that the most visually noticeable sinusoidal oscillations, which seem to be the main part of the EEG, are in fact associated not with the transmission of information, but with the regulation of the state of brain tissue.

I only believe in mechanistic explanations. Your view is insightful, and I agree to most of them.
My mentor has a very simple and mechanistic view, at least at a large spatial scale, that unconscious brain = low-frequency dominant, strongly oscillatory and coherent everywhere, while awake, cognitively busy brain = more high freqs, less oscillatory, less coherent. This view is simple and reliable. Hopefully we can generalize this safe image to other brain states without running too much risk.

After all, as you say, a clear behavioral criterion i.e., eyes open-closed should be used to identify at least a clear subset of 'alpha' so that we can start from the most robust definition. Every time I hear people talk about alpha as a known phenomenon, it drives me nuts. I have little tolerance to violation of conceptual clarity.

Can we even find as popular, dominant, yet confused equivalent in other fields than EEG alpha?

[MarjanZamani1990]

@MakotoMiyakoshi

Thank you for your response.
I realize I may have misinterpreted your previous message. When you mentioned group organization and openness to suggestions, I took it as an invitation to provide structured input—apologies if my response came across as overly prescriptive.
I completely respect that this is your work and led by you, and I did not intend to interfere with its direction.
I remain very interested in the scientific aspects of this topic, particularly the generative mechanisms of aperiodic activity and their relation to empirical data.
If you have time, I would really appreciate any critical feedback on my ideas and approach. It would be very valuable for me as I continue developing my research skills.

[MakotoMiyakoshi]

Hi @MarjanZamani1990,

When you mentioned group organization and openness to suggestions, I took it as an invitation to provide structured input

Yes, I do want to organize something to do involving as many as possible as a festival. But I'm not ready yet.

I remain very interested in the scientific aspects of this topic, particularly the generative mechanisms of aperiodic activity and their relation to empirical data.

So this is a part of a problem: when I become curious about something, say 1/f^alpha-ness, I do literature search, I relate my ongoing studies to the topic, etc.. Then I learn many things, which explain what it is. The more I understand, the less I become curious because I'm now learned and less mystery remains!

However, I also want to share what I learned with others who are less enthusiastic about performing deep literature search. I totally understand it, and it is not due to the lack of enthusiasm but it is because they do not know how to do it meaningfully.

So, the easiest way for me to share my understanding is to have an online/offline meeting for discussion. If I want to do it most properly, when it'll take a form of publication.

Your main interest seems to be in signal processing. Unfortunately, I'm less counting on that approach in this topic. I'm more interested in the generative mechanism rather than post processing. But I'm sure there are people who love signal processing like you do in this community. I encourage you to look for someone like that.

[JazminMorrone]

Hello,

It looks like the Zoom recording link is no longer active. Would it be possible to update the link and/or upload the presentations from the meeting to an accessible location?

Thank you in advance!

[MakotoMiyakoshi]

Hi @JazminMorrone,

Thank you for your interest and sorry for inconvenience!
I moved the recording file back from trash to online (after 30 days, it was auto-trashed).
Here is the URL to access the video. Please let me know if you still can't watch it.

https://ucsd.zoom.us/rec/share/jQV2m64c6FnNot0HX2TY0nm3NkYyr3Q-Ntr-Vtplj8sKjAnQ0xBYdYgC_GJtp1Mc.h8l8Z8t3NsTU7iAB?startTime=1770998422000
Passcode: =0VP&M+p

[A-Mirifar]

Dear Makoto, unfortunately, the link doesn’t work. Could you please check to see what might be wrong, thank you in advance!

[MakotoMiyakoshi]

Hi @A-Mirifar,

The Zoom service keeps it only for 30 days and automatically moved it to trash box. It seems I can still recover it though.
https://ucsd.zoom.us/recording/detail?meeting_id=lrhD9qbIRNCuhIuH62EPXA%3D%3D

Let me know if you can download the video files.

[A-Mirifar]

Dear Makoto, the original link is working now, and the recorded video is available again. Thanks!

[MakotoMiyakoshi]

HI @A-Mirifar, Good to confirm that!

[EugenMasherov]

It's wonderful when understanding increases as knowledge is gained. But sometimes the effect is the opposite: the complexities of incomprehensible facts grow as information accumulates, like the surface area of a balloon expanding when it's inflated. As for signal processing, it's closer to my heart as a field of research, but the research tool must be appropriate to the object of study. Therefore, it's necessary to first understand the signal generation mechanism and then develop a method for processing them. The alternative approach—taking an existing method successfully used in another field and mechanically applying it to your own—is significantly simpler, but can lead to self-deception. вт, 19 мая 2026 г. в 04:17, Makoto Miyakoshi ***@***.***>:

…

Hi @MarjanZamani1990 <https://github.com/MarjanZamani1990>, When you mentioned group organization and openness to suggestions, I took it as an invitation to provide structured input Yes, I do want to organize something to do involving as many as possible as a festival. But I'm not ready yet. I remain very interested in the scientific aspects of this topic, particularly the generative mechanisms of aperiodic activity and their relation to empirical data. So this is a part of a problem: when I become curious about something, say 1/f^alpha-ness, I do literature search, I relate my ongoing studies to the topic, etc.. Then I learn many things, which explain what it is. The more I understand, the less I become curious because I'm now learned and less mystery remains! However, I also want to share what I learned with others who are less enthusiastic about performing deep literature search. I totally understand it, and it is not due to the lack of enthusiasm but it is because they do not know how to do it meaningfully. So, the easiest way for me to share my understanding is to have an online/offline meeting for discussion. If I want to do it most properly, when it'll take a form of publication. Your main interest seems to be in signal processing. Unfortunately, I'm less counting on that approach in this topic. I'm more interested in the generative mechanism rather than post processing. But I'm sure there are people who love signal processing like you do in this community. I encourage you to look for someone like that. — Reply to this email directly, view it on GitHub <#15?email_source=notifications&email_token=B2DLTZBE2N6GMMIE3EW3O6343OYYZA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNRZGY4TCOBTUZZGKYLTN5XKO3LFNZ2GS33OUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-16969183>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/B2DLTZE7LMNCYU4J6JMTNEL43OYYZAVCNFSM6AAAAACU2DLMRSVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMOJWHEYTQMY> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[MakotoMiyakoshi]

Hi @EugenMasherov,

it's necessary to first understand the signal generation mechanism

Yes, namely the forward problem.