MontageFlow: Physically Valid High Fidelity EEG Generation

Sudarshan Regmi; Nabaraj Subedi; Arvind Pillai; Yu Yvonne Wu; Yuliang Chen; Michael V. Heinz; Lisa Marsch; Nicholas C. Jacobson

cs.AIOct 10, 2026

MontageFlow: Physically Valid High Fidelity EEG Generation

Authors: Sudarshan Regmi, Nabaraj Subedi, Arvind Pillai, Yu Yvonne Wu, Yuliang Chen, Michael V. Heinz, Lisa Marsch, Nicholas C. Jacobson, +1 more

Organizations: Dartmouth College

Abstract

Generative modeling offers a promising solution to the scarcity of labeled clinical electroencephalography (EEG) data under strict privacy constraints. Although recent generative models report substantial improvements in generation fidelity, we show that these gains conceal a categorical failure. EEG is not an arbitrary multichannel signal: each channel records the voltage difference between two scalp electrodes. So, real recordings satisfy exact linear identities among their channels. Current generators invariably violate these identities, yet this has gone unnoticed. We therefore propose MontageFlow, a flow-matching model that enforces these identities by restricting its output to signals that satisfy them. We find that ensuring validity improves fidelity too. Furthermore, to retain the EEG characteristics that oversmoothing washes out, we also condition the model on a physiological state vector encoded from the whole EEG window. For generation, this vector is drawn from a prior fitted in latent space. Across 3 large-scale benchmarks, our model improves TS-FID by 1.6-3.6 $\times$ over the closest rival while ensuring validity.

Figures & tables

Figure 1: The bipolar montage induces two exact linear constraints. (a) Each channel is an oriented difference between two electrode potentials. Channels 1–4 and 9–12 form two paths from Fp1 to O1, and channels 5–8 and 13–16 do the same from Fp2 to O2. Either path sums to the potential difference between its endpoints. (b) The two left path sums overlap on a real TUSZ window, and (c) their difference is numerically zero.

Figure 2: Overview of MontageFlow. During training, an encoder

f_{\phi}

maps an EEG window

X

to a global latent

h

. The decoder receives the current flow state

X_{t}

, flow time

t

, class label

y

, and perturbed latent

\widetilde{h}

, and predicts a velocity field. After encoder-decoder training, a class-conditional rectified-flow prior

p_{\psi}(h\mid y)

is fitted to the encoded training latents and replaces the encoder during generation. The predicted velocity is projected onto

\ker A

. At inference, the Gaussian source is also projected to guarantee valid EEG.

	classes	window	train windows	test windows
TUSZ	2	$16\times 1000$	25,542	9,538
TUEV	6	$16\times 1000$	68,445	29,421
TUAB	2	$16\times 2000$	297,103	36,945

Table 1: Corpora used in our study. Windows lie on the 16-channel bipolar montage at 200 Hz.

Method	TS-FID (train) $\downarrow$			TS-FID (test) $\downarrow$			montage index (test) $\downarrow$
Method	TUSZ	TUEV	TUAB	TUSZ	TUEV	TUAB	TUSZ	TUEV	TUAB
EEG-GAN	$342.3$	$316.9$	$394.5$	$331.4$	$400.9$	$455.7$	$0.539$	$0.519$	$0.505$
Vanilla diffusion	$113.2$	$82.4$	$120.1$	$121.6$	$145.8$	$149.1$	$\underline{0.069}$	$\underline{0.078}$	$\underline{0.071}$
Vanilla flow matching	$125.0$	$123.9$	$157.1$	$137.0$	$212.8$	$186.0$	$0.283$	$0.437$	$0.294$
JET	$207.5$	$212.5$	$244.6$	$199.1$	$301.9$	$285.8$	$0.991$	$0.988$	$0.986$
JET ×	$203.2$	$139.0$	$211.5$	$198.5$	$221.9$	$252.6$	$0.151$	$0.217$	$0.148$

Table 2: Generation quality and physical validity ( bold : best, underline : second best).

Method	TUSZ	TUEV	TUAB
EEG-GAN	$0.412$	$0.674$	$0.668$
Vanilla diffusion	$0.928$	$0.948$	$0.961$
Vanilla flow matching	$0.926$	$0.935$	$0.963$
JET	$-0.033$	$0.215$	$0.024$
JET ×	$0.916$	$0.908$	$0.948$
MontageDiff	$\underline{0.948}$	$\underline{0.973}$	$\underline{0.981}$

Table 3: Shape correlation on the test split (

\uparrow

, bold : best, underline : second best).

Method	TUSZ	TUEV	TUAB	Avg.
Real recordings	$0.810$	$0.627$	$0.797$	$0.745$
EEG-GAN	$0.499$	$0.197$	$0.502$	$0.399$
Vanilla diffusion	$0.580$	$\mathbf{0.607}$	$0.709$	$0.632$
Vanilla flow matching	$0.576$	$0.396$	$0.592$	$0.521$
JET	$\underline{0.750}$	$0.401$	$0.581$	$0.577$
JET ×	$0.663$	$0.404$	$0.623$	$0.563$

Table 4: Downstream balanced accuracy (

\uparrow

, bold : best, underline : second best).

Method	TUSZ		TUEV		TUAB
Method	train	test	train	test	train	test
Vanilla flow matching ( $x$ -parameterisation)	$182.9$	$180.4$	$197.8$	$276.1$	$217.1$	$246.0$
$+$ velocity parameterisation	$136.9$	$138.2$	$150.7$	$233.9$	$181.5$	$215.2$
$+$ cross-channel attention	$125.0$	$137.0$	$123.9$	$212.8$	$157.1$	$186.0$
$+$ latent conditioning	$60.4$	$74.6$	$46.8$	$120.6$	$86.5$	$115.1$
$+$ latent perturbation	$49.8$	$60.8$	$31.8$	$106.3$	$81.7$	$110.5$

Table 5: Incremental ablation from vanilla flow matching to MontageFlow (TS-FID

\downarrow

Projection	TS-FID $\downarrow$	PSD slope $\downarrow$	Envelope $\downarrow$	Hj. act. $\downarrow$	Hj. mob. $\downarrow$	Drift $\downarrow$
Inference only	140.6	0.629	0.244	0.256	0.366	0.097
Training $+$ inference	59.9	0.400	0.164	0.215	0.232	0.092

Table 6: Is post-hoc projection enough? TUAB test split (all corpora: Table 9 ). Inference only projects the model of row 5 of Table 5 at sampling time, and training

+

inference is MontageFlow ( bold : best).

Figure 3: Latent conditioning mitigates oversmoothing. Representative samples from TUAB are shown.

Method	Patch	Params	s/step	GPU-h	ms/window
EEG-GAN	–	130M	2.40	36.0	2
Vanilla diffusion	100	130M	0.17	7.5	73
Vanilla flow matching	100	130M	0.17	7.5	74
JET	200	130M	0.11	5.0	40
JET ×	200	130M	0.08	3.7	37
MontageDiff	100	156M	0.23	10.6	76

Table 7: Computational cost across methods on TUAB.

Figure 4: Physiological diagnostics. Top: power spectrum, median over windows, with shading from the 10th to the 90th percentile. Middle: amplitude over time (root-mean-square across channels), median over windows, with shading over the middle half of windows. Bottom: distribution of sample amplitudes (log density). Windows are 10 s on TUAB and 5 s on TUEV and TUSZ.

Method	TS-FID $\downarrow$	PSD slope $\downarrow$	Envelope $\downarrow$	Hj. act. $\downarrow$	Hj. mob. $\downarrow$	Drift $\downarrow$	Avg. $\downarrow$
EEG-GAN	331.4	1.846	1.216	1.362	0.831	0.712	1.193
Vanilla diffusion	121.6	0.684	1.679	1.652	0.571	0.204	0.958
Vanilla flow matching	137.0	0.199	1.651	1.585	0.391	0.102	0.786
JET	199.1	0.862	0.615	0.506	0.673	0.410	0.613
JET ×	198.5	0.897	0.518	0.467	0.694	0.282	0.571
MontageDiff	40.6	1.141	0.515	0.587	0.480	0.065	0.557

Table 8: Physiological diagnostics on the TUSZ test split ( bold : best, underline : second best).

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 5: Generated and real windows. First four channels of the bipolar montage, five seconds, one window per corpus and method (the median-amplitude window among

2{,}000

samples).

Figure 6: Channel-wise power spectrum. Median over windows for each channel, with shading from the 10th to the 90th percentile.

Figure 7: Channel-wise amplitude over time. Absolute amplitude of each channel, median over windows, with shading over the middle half of windows.

Figure 8: Channel-wise amplitude distribution. Distribution of sample amplitudes for each channel (log density).

Corpus	Projection	TS-FID $\downarrow$	PSD slope $\downarrow$	Envelope $\downarrow$	Hj. act. $\downarrow$	Hj. mob. $\downarrow$	Drift $\downarrow$
	Inference only	63.1	0.257	0.639	0.584	0.471	0.089
TUSZ	Training $+$ inference	44.0	0.221	0.580	0.551	0.302	0.069
	Inference only	107.6	0.268	0.267	0.280	0.217	0.129
TUEV	Training $+$ inference	93.1	0.230	0.261	0.278	0.210	0.121
	Inference only	140.6	0.629	0.244	0.256	0.366	0.097
TUAB	Training $+$ inference	59.9	0.400	0.164	0.215	0.232	0.092

Appendix

Table 9: Post-hoc projection vs. projection in training on all three corpora, test split ( bold : best per corpus).

Explore similar work

Let EEG Models Learn EEG
May 20, 2026Yifan Wang, Yijia Ma, Wen Li +1Electroencephalography Time Series Generation
Stony Brook University · University of Texas Health Center at Houston.
Not All EEG Moments Are Equal: Position-Adaptive Time Scheduling for EEG Generation
Jul 25, 2026Boheng Liu, Ziyu Li, Chenghua Duan +2Brain-Computer Interfaces Time Series Generation
School of Computer Science and Technology Beijing Institute of Technology Beijing, China
B[FM] $^2$ : Brain Foundation Model via Flow Matching with SplitUNet
Jun 18, 2026Jaedong Hwang, Kathleen Zhang, Wei Dai +5Flow Matching EEG Decoding
Massachusetts Institute of Technology · KU Leuven

May 20, 2026J/K move · Enter open · S save

Let EEG Models Learn EEG

Yifan Wang, Yijia Ma, Wen Li, Chenyu You

Stony Brook University · University of Texas Health Center at Houston.

Electroencephalography Time Series Generation Conditional Generative Modeling

Open paper AskPDF arXiv

High-fidelity EEG generation is critical for alleviating data scarcity and addressing privacy constraints in large-scale neural modeling. Despite recent progress, most existing approaches formulate EEG generation via discrete denoising objectives, which inadequately reflect the inherently continuous temporal dynamics and spectral structure of neural activity. As a result, these methods often struggle to preserve long-range temporal dependencies and exhibit mismatches in the spectral and temporal structure of the generated signals. In this work, we argue that effective EEG generation requires models that operate directly on the continuous evolution of neural signals. We introduce Just EEG Transformer (JET), a generative framework based on conditional flow matching that models EEG as raw sequences evolving along continuous trajectories. By learning a smooth vector field that transports noise to the EEG data distribution, JET captures temporal continuity and transient dynamics without relying on discretized denoising schemes or domain-specific representations. To ensure that the learned dynamics remain consistent with key properties of EEG signals, we introduce principled constraints that preserve spectral structure, temporal stationarity, and signal-level statistics. Across three large-scale benchmarks, JET consistently achieves state-of-the-art performance, reducing TS-FID by over 40% compared to strong baselines. Extensive analyses show that JET captures key structural properties of neural dynamics, providing a scalable and principled approach to EEG generation. Project page: https://y-research-sbu.github.io/JET/ .

Abstract

Explore similar work

Let EEG Models Learn EEG

Not All EEG Moments Are Equal: Position-Adaptive Time Scheduling for EEG Generation

B[FM]2^22: Brain Foundation Model via Flow Matching with SplitUNet

B[FM] $^2$ : Brain Foundation Model via Flow Matching with SplitUNet