- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper introduces a geometry-informed Gaussian source distribution for flow-matching models of multichannel brain signals. It constructs a nearest-neighbor graph from sensor coordinates and defines the source covariance through a Matérn function of the normalized graph Laplacian, while retaining the original training objective, coupling procedure, and drift architecture. Across eight EEG datasets and four flow-matching formulations, the method reports a 12 to 17% geometric-mean reduction in PSD-KL, with additional evaluations on MEG, intracranial EEG, traffic signals, and EEG classification augmentation. Ablations examine the contributions of the spatial eigenbasis, covariance spectrum, and graph construction, suggesting that alignment with local sensor geometry is an important factor in the observed improvements.
Critical strengths
- The authors introduce a physically motivated spatial bias through the source covariance without adding learned parameters or changing the flow-matching objective.
- Constructing this prior from coordinates alone and normalizing its total variance makes the intervention straightforward to implement and reasonably well controlled.
- The authors conducted comprehensive experiments comparing to multiple flow matching methods to showcase the good performance.
Critical weaknesses
- The comparison of this paper mainly focused on the flow matching baselines, but I think as a work about Multi-Channel Brain Signals generation, the authors should conduct comparison directly with the related works like JET, GPT2MEG, MEG-GPT, EEG-GAN etc. It can better showcase the improvement of generated signals.
- In the paper the authors claim that All methods share a 1D U-Net drift, but recent works show that transformer based backbone can have better performance. Can authors conduct simple experiments based on transformer backbone?
- The downstream evaluation is restricted to Mumtaz-MDD, the authors should provide more donstream evaluation results across different datasets.
- It seems that the major improvement of downstream task comes from adding iso-source, the graph-prior only gives limited improvement. The author should have further analysis to show that this prior is pratically usefull.
Questions for authors
Please refer to the weakness.
4 comments
Response to Reviewer cH53 (2/2)
Re: The downstream evaluation is restricted to Mumtaz-MDD, the authors should provide more donstream evaluation results across different datasets. Re: It seems that the major improvement of downstream task comes from adding iso-source, the graph-prior only gives limited improvement. The author should have further analysis to show that this prior is pratically usefull.
We agree with the reviewer's point that the improvement in Table 5 comes from adding generated data. The graph prior is slightly higher than the isotropic prior ( vs. ).
We ran the same experiment on the remaining seven datasets (Table cH53-3). Augmentation improves accuracy only on Mumtaz-MDD. It makes no difference on TUAB, TUEV, and SEED-V, and it decreases accuracy on BCI-IV 2a, FACED, SHU, and PhysioNet-MI, for both priors. When generated data from the isotropic prior does not help, generated data from the graph prior does not help either.
The graph prior is designed to improve generation quality. On the datasets with more channels, it lowers PSD-KL for all four methods ( on SEED-V, on PhysioNet-MI) and improves TS-FID and the autocorrelation error (Tables cH53-4 and cH53-5).
Table cH53-3. Downstream augmentation on each dataset with an EEGNet-8,2 classifier.
| Dataset(ch) | Real only | +iso-source | +graph-prior |
|---|---|---|---|
| TUAB(16) | 0.769±0.030 | 0.764±0.023 | 0.751±0.026 |
| TUEV(16) | 0.462±0.025 | 0.462±0.049 | 0.453±0.031 |
| Mumtaz-MDD(19) | 0.608±0.146 | 0.819±0.037 | 0.833±0.054 |
| BCI-IV 2a(22) | 0.563±0.032 | 0.374±0.028 | 0.388±0.031 |
| FACED(32) | 0.182±0.027 | 0.132±0.027 | 0.140±0.035 |
| SHU(32) | 0.831±0.009 | 0.816±0.007 | 0.807±0.014 |
| SEED-V(62) | 0.370±0.023 | 0.356±0.027 | 0.363±0.032 |
| PhysioNet-MI(64) | 0.586±0.013 | 0.536±0.022 | 0.502±0.013 |
Table cH53-4. TS-FID (lower is better) for the four methods with the isotropic source and with the graph-Matérn source (GP-). Bold denotes the better source within each method.
| Dataset(ch) | SF2M | GP-SF2M | SI | GP-SI | OT-CFM | GP-OT-CFM | RF | GP-RF |
|---|---|---|---|---|---|---|---|---|
| TUAB(16) | 473.17±11.66 | 512.52±34.27 | 473.09±22.31 | 514.74±19.93 | 468.35±25.85 | 510.91±19.22 | 590.37±79.91 | 624.72±93.63 |
| TUEV(16) | 658.04±22.40 | 641.44±37.19 | 651.20±19.27 | 647.42±40.13 | 652.32±29.64 | 658.51±32.15 | 664.17±59.77 | 647.29±27.79 |
| Mumtaz-MDD(19) | 514.88±22.17 | 530.41±16.29 | 497.56±47.23 | 509.19±42.41 | 492.32±30.10 | 517.56±13.95 | 522.95±28.13 | 548.31±23.80 |
| BCI-IV 2a(22) | 158.68±15.31 | 180.03±28.93 | 178.24±8.54 | 176.22±14.87 | 170.25±29.03 | 176.08±23.68 | 180.68±23.19 | 183.23±33.82 |
| FACED(32) | 313.52±10.06 | 284.54±25.52 | 307.77±17.40 | 301.14±46.48 | 302.73±17.05 | 310.85±58.38 | 314.23±26.85 | 293.09±15.81 |
| SHU(32) | 728.66±99.82 | 520.43±62.73 | 619.76±86.87 | 531.10±53.68 | 701.91±69.70 | 550.46±101.70 | 767.51±75.19 | 553.10±45.42 |
| SEED-V(62) | 171.01±22.91 | 81.01±9.85 | 155.52±15.21 | 87.40±4.52 | 173.85±8.28 | 76.19±8.93 | 152.03±34.84 | 84.86±15.62 |
| PhysioNet-MI(64) | 570.35±58.08 | 277.21±54.57 | 585.62±69.31 | 287.35±62.64 | 537.23±103.35 | 300.79±57.54 | 612.13±59.51 | 323.20±62.61 |
Table cH53-5. Autocorrelation error, multiplied by 100 (lower is better).
| Dataset(ch) | SF2M | GP-SF2M | SI | GP-SI | OT-CFM | GP-OT-CFM | RF | GP-RF |
|---|---|---|---|---|---|---|---|---|
| TUAB(16) | 0.68±0.23 | 0.81±0.18 | 0.56±0.18 | 0.84±0.32 | 0.73±0.31 | 0.81±0.26 | 1.27±0.29 | 1.21±0.31 |
| TUEV(16) | 9.30±6.84 | 7.47±2.12 | 7.14±2.72 | 6.18±1.63 | 7.36±1.42 | 7.76±2.04 | 10.64±1.97 | 8.80±1.87 |
| Mumtaz-MDD(19) | 0.95±0.21 | 1.12±0.35 | 0.88±0.34 | 0.95±0.44 | 0.92±0.42 | 1.10±0.24 | 0.95±0.21 | 1.01±0.38 |
| BCI-IV 2a(22) | 3.97±0.69 | 4.49±1.03 | 4.46±0.31 | 4.42±0.53 | 4.30±0.81 | 4.42±0.65 | 4.51±0.52 | 4.52±1.08 |
| FACED(32) | 5.59±0.54 | 3.54±1.74 | 5.35±1.12 | 3.85±1.54 | 4.86±1.00 | 4.18±0.87 | 5.02±0.85 | 4.56±0.60 |
| SHU(32) | 24.68±4.46 | 22.80±1.67 | 23.13±2.79 | 21.62±2.67 | 25.14±1.77 | 20.87±7.25 | 26.55±1.65 | 23.02±1.29 |
| SEED-V(62) | 7.35±0.09 | 6.46±0.27 | 7.29±0.14 | 6.67±0.12 | 7.39±0.02 | 6.37±0.20 | 7.25±0.20 | 6.51±0.36 |
| PhysioNet-MI(64) | 5.84±0.57 | 4.25±1.12 | 5.98±0.76 | 4.46±1.33 | 5.06±1.49 | 4.39±1.35 | 6.75±0.52 | 5.11±0.88 |
I sincerely thank the authors for the additional experiments clarification; they address my concerns about baselines and backbones. However, since the gains are largely confined to high-channel montages and the downstream benefit does not generalize, I will maintain my current score.
We appreciate the reviewer's response.
Re: The gains are largely confined to high-channel montages
We agree that the gain is concentrated on the datasets with more channels. When the sensors are dense, neighboring channels are strongly correlated. The isotropic source ignores this correlation, while the graph prior starts from noise that already has it. Dense montages are therefore the setting in which the choice of source matters, and there the gain is consistent. On SEED-V and PhysioNet-MI the graph prior lowers PSD-KL for all four methods, by 15% and 33 to 41% with the U-Net (Table 2) and by 19% and 30 to 36% with the transformer (Table cH53-2).
Re: The downstream benefit does not generalize
Downstream accuracy is a useful proxy for sample quality, but more realistic samples do not always give better augmentation. On ImageNet, adding nearly photorealistic generated images to the training set improves top-5 accuracy by about 0.2% in the best case and otherwise degrades it, and FID does not predict classification accuracy [R1]. Samples from a good diffusion model can even harm contrastive learning [R2]. We observe the same. Generated data do not raise accuracy on seven of the eight datasets with either source (Table cH53-3). Our claim is about generation quality (PSD-KL, wPLI, TS-FID, and autocorrelation error).
References
[R1] Ravuri and Vinyals. Classification Accuracy Score for Conditional Generative Models. NeurIPS. 2019.
[R2] Wang, Zhang, and Wang. Do Generated Data Always Help Contrastive Learning? ICLR. 2024.
Response to Reviewer cH53 (1/2)
We appreciate the reviewer's constructive feedback and address each concern below.
Re: The comparison of this paper mainly focused on the flow matching baselines, but I think as a work about Multi-Channel Brain Signals generation, the authors should conduct comparison directly with the related works like JET, GPT2MEG, MEG-GPT, EEG-GAN etc. It can better showcase the improvement of generated signals.
We trained EEGDM and JET on each of the eight datasets with their official codebases and default settings, with five random seeds, and evaluated them with the metrics of Tables 2 and 3 (Table cH53-1). EEGDM, which denoises one channel at a time, reaches a PSD-KL between 1.43 (PhysioNet-MI) and 17.94 (BCI-IV 2a), with standard deviations below 1 on every dataset. JET reaches 3.29 on TUAB and 6.59 on TUEV, the two datasets its recipe was developed on. On the other six its PSD-KL is between 23.11 and 1662.54, and its seeds differ widely (for example 43 to 546 on SEED-V). Samples of both models carry almost no coupling between channels. Their wPLI correlation is between -0.12 and 0.05 on every dataset.
Table cH53-1. EEGDM and JET, trained on our training splits with their released code and default settings and scored on Table 2's test windows. PSD-KL (lower is better) as in Table 2 and wPLI correlation (higher is better) as in Table 3, mean ± std over five random seeds.
Re: In the paper the authors claim that All methods share a 1D U-Net drift, but recent works show that transformer based backbone can have better performance. Can authors conduct simple experiments based on transformer backbone?
We appreciate the reviewer's suggestion. We tested a DiT-style transformer (patch 8, width 256, depth 8) with four flow-matching methods used in Table 2 (Table cH53-2). On the dense montages the pattern is the same as with the U-Net. The graph source lowers PSD-KL by 21 to 23% on SHU, by 19% on SEED-V and by 30 to 36% on PhysioNet-MI for SF2M, SI, OT-CFM and RF. On the five sparser montages the differences are mostly within seed variance, again as with the U-Net. At this size the transformer is the weaker generator, with a higher PSD-KL than the U-Net in 61 of the 64 cells (eight datasets, four methods, two sources; for SI on FACED, for example, 822 against 33), so the U-Net remains our main model. Because the gain does not appear on every dataset with either network, we will change "works with any coupling and any drift network" in the abstract to say that the prior requires no change to the coupling or the drift network.
Table cH53-2. Transformer drift in place of the U-Net, everything else as in Table 2, in Table 2's layout. PSD-KL, mean ± sample SD over five seeds (lower is better), for the four methods with the isotropic source and with the graph-Matérn source (GP-). Bold denotes the better source within each method.