- Rating
- 3
- Confidence
- 3
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
NeurDuo-EEG combines a fast per-channel state-space stream, periodic cross-channel attention, a slower memory-writing stream, and bounded explicit memory. It is pretrained on approximately 3,955 hours of EEG and evaluated on short-window tasks as well as sleep staging and seizure detection with extended causal context. The paper compares model sizes, context lengths, architectural components, and carrying versus resetting recurrent state. The sequence evaluation reports AUC-PR, sleep-staging metrics, and false alarms per hour, alongside inference latency and memory measurements.
Critical strengths
-
Direct evidence that persistent backbone state is useful. With the same evaluation framework, carrying rather than resetting state improves CHB-MIT 15-channel AUC-PR from 0.307 to 0.413 and Sleep-EDF macro-F1 from 0.745 to 0.755 (Table 6). The context-length sweep is also informative because it reports non-monotonic behavior rather than implying that longer context always helps. The Large model reaches 0.471 AUC-PR on the 15-channel seizure task, compared with 0.285 for the strongest reported classical pipeline (Table 3).
-
Unusually thorough protocol analysis and disclosure. Architectural variants are re-pretrained, the common readout is capacity controlled, and both short-window and sequence tasks are considered. The appendix identifies relevant baseline pretraining overlaps and the shared chb01/chb21 patient, and quantifies the effect of removing that patient's test units. This sensitivity analysis preserves the AUC-PR advantage. These details make the experimental evidence substantially more interpretable.
Critical weaknesses
-
The streaming-cost comparison does not measure the baseline pipelines used for the accuracy results. Section 5.3 and Appendix F.2 compare advancing NeurDuo by 0.5 seconds with re-encoding the entire history for each baseline. However, the sequence-task accuracy protocol encodes baseline inputs in separate 30-second units and uses a causal Mamba context head. Such a pipeline can cache unit embeddings and recurrent head state; it need not repeatedly encode an hour of raw EEG. The current experiment demonstrates the advantage over an expanding-window implementation, but does not establish the deployment advantage over those practical baselines. In addition, the 4.7M Small-model figure is a backbone count: the 15-channel CHB-MIT readout contains another 24.81–24.95M parameters (Table 16). Please report complete-pipeline latency, memory, and parameter counts at a common prediction cadence, including the readout and required signal accumulation time. This would preserve the valid bounded-state result while making the comparison relevant to the reported task performance.
-
The false-alarm statistic has a boundary artifact and is not a deployment estimate. Appendix C explicitly sets the threshold from the evaluated fold's positive-unit scores and merges consecutive positive predictions across recording boundaries (p. 24, lines 1279–1288). The threshold choice is transparently described as a common-sensitivity comparison, which is acceptable as an exploratory operating-point analysis. Merging across separate recordings, however, can join unrelated alarms or attach a false alarm to a seizure in another record. Please recompute alarms with strict recording and participant boundaries. A validation-selected threshold, held fixed on test data, together with event-level sensitivity and detection delay, is needed if the paper retains practical alarm-performance claims. This concern does not invalidate the threshold-independent AUC-PR comparison, which is the stronger part of the evidence.
Questions for authors
1.How does NeurDuo compare with cached 30-second baseline encoders plus the existing causal context head when measuring the entire inference pipeline at the same output cadence? Please include the task-head parameters and accumulation latency.
2.How much do FP/h results change after preventing merges across recordings? Can the authors additionally report test results using thresholds selected only on validation subjects, with event sensitivity and delay?
3.Table 5 gives CHB-MIT AUC-PR of 0.413 for the full model, 0.389 without retrieval, and 0.406 for the Mamba-only variant. What are the paired fold-level differences? The evidence for persistent state is stronger than the evidence that the explicit-memory construction itself explains the long-sequence gain. Clarifying this distinction would materially improve the contribution claim.
5 comments
[Q2] How much do FP/h results change after preventing merges across recordings? Can the authors additionally report test results using thresholds selected only on validation subjects, with event sensitivity and delay?
Merges across recordings. Stopping every alarm at the boundary of its recording changes FP·h⁻¹ very little:
- At the operating point of Table 3, the change is at most 1.5 % for any model at either montage, and at most 0.2 % for the NeurDuo models.
- At the validation-selected operating point below, the change is at most 1.3 % (Tables R2 and R3 report both counts).
- No ranking changes. In the revision, every FP·h⁻¹ is counted within recordings.
Validation-selected thresholds. Tables R2 and R3 report the operating point chosen without test labels:
- For each fold, the threshold that reaches 50 % unit sensitivity on the validation subjects, at the selected epoch, is applied unchanged to the test subjects.
- Within each recording, consecutive above-threshold 30 s units form one alarm. Consecutive seizure units form one seizure, which counts as detected if any alarm overlaps it.
- Detection delay runs from seizure onset to the end of the first above-threshold unit inside the seizure. The shortest possible delay is therefore one unit, 30 s.
- Counts and hours are pooled over the five folds.
- The submitted runs did not store their validation predictions, so every model was re-trained once with the same folds and seed.
- The tables include CodeBrain (ICLR 2026), a recent EEG foundation model we added as a baseline in this revision, evaluated under the same protocol.
Table R2 · CHB-MIT, 15 channels, threshold selected on the validation subjects.
| Method | Pre-trained | FP·h⁻¹ | FP·h⁻¹, alarms merged across recordings | Unit sensitivity | Event sensitivity | Median delay (s) |
|---|---|---|---|---|---|---|
| BP-GBDT | — | 1.82 | 1.81 | 0.42 | 0.55 (101/185) | 30 |
| FBCov-TS-Lin | — | 3.58 | 3.54 | 0.43 | 0.46 (86/185) | 30 |
| ERP-Lin | — | 15.49 | 15.31 | 0.50 | 0.67 (124/185) | 30 |
| BIOT | yes | 1.45 | 1.45 | 0.37 | 0.44 (82/185) | 30 |
| CBraMod | yes | 4.45 | 4.43 | 0.34 | 0.33 (61/185) | 30 |
| LaBraM | yes | 9.81 | 9.78 | 0.32 | 0.41 (75/185) | 30 |
| EEGPT | yes | 6.70 | 6.68 | 0.31 | 0.35 (65/185) | 30 |
| ST-EEGFormer-S | yes | 23.29 | 23.20 | 0.41 | 0.66 (122/185) | 30 |
| REVE-Base | yes | 6.54 | 6.52 | 0.37 | 0.55 (102/185) | 30 |
| CodeBrain | yes | 2.39 | 2.38 | 0.37 | 0.41 (75/185) | 30 |
| NeurDuo-Small | yes | 0.28 | 0.28 | 0.39 | 0.43 (79/185) | 60 |
| NeurDuo-Base | yes | 1.00 | 0.99 | 0.43 | 0.47 (87/185) | 30 |
| NeurDuo-Large | yes | 1.25 | 1.25 | 0.39 | 0.52 (97/185) | 30 |
Table R3 · CHB-MIT, 2 channels (T7, P8), same protocol.
| Method | Pre-trained | FP·h⁻¹ | FP·h⁻¹, alarms merged across recordings | Unit sensitivity | Event sensitivity | Median delay (s) |
|---|---|---|---|---|---|---|
| BP-GBDT | — | 1.20 | 1.20 | 0.34 | 0.42 (78/185) | 30 |
| FBCov-TS-Lin | — | 2.61 | 2.61 | 0.39 | 0.52 (96/185) | 30 |
| ERP-Lin | — | 12.89 | 12.73 | 0.55 | 0.70 (130/185) | 30 |
| BIOT | yes | 5.87 | 5.85 | 0.40 | 0.52 (97/185) | 30 |
| CBraMod | yes | 1.69 | 1.69 | 0.36 | 0.44 (81/185) | 30 |
| LaBraM | yes | 6.56 | 6.49 | 0.37 | 0.50 (92/185) | 30 |
| EEGPT | yes | 2.99 | 2.96 | 0.36 | 0.44 (82/185) | 30 |
| ST-EEGFormer-S | yes | 14.24 | 14.20 | 0.44 | 0.64 (118/185) | 30 |
| REVE-Base | yes | 12.10 | 12.05 | 0.48 | 0.69 (127/185) | 30 |
| CodeBrain | yes | 3.20 | 3.19 | 0.40 | 0.46 (85/185) | 30 |
| NeurDuo-Small | yes | 0.27 | 0.27 | 0.46 | 0.48 (88/185) | 30 |
| NeurDuo-Base | yes | 1.11 | 1.11 | 0.49 | 0.55 (101/185) | 30 |
| NeurDuo-Large | yes | 0.54 | 0.54 | 0.46 | 0.45 (83/185) | 30 |
Results:
- NeurDuo raises the fewest false alarms of all models, at both montages.
- At 15 channels, the three sizes raise 0.28–1.25 false alarms per hour, against at least 1.45 for every baseline.
- At 2 channels, they raise 0.27–1.11, against at least 1.20. Their unit sensitivity (0.46–0.49) is higher than that of every baseline except ERP-Lin and REVE-Base, which raise 12–13 false alarms per hour.
- This holds at useful event sensitivity. NeurDuo detects 0.43–0.55 of the seizures. The pre-trained models that detect more, ST-EEGFormer-S and REVE-Base (0.55–0.69), pay for it with 6.5–23.3 false alarms per hour, 5 to 80 times as many.
- Delay. The median delay is one 30 s unit, the shortest possible, for every model except NeurDuo-Small at 15 channels (two units). That model runs at the lowest false-alarm rate of all, 0.28 per hour.
We will make corresponding revisions to the claim in the main text:
- All FP·h⁻¹ values are counted within recordings.
- Tables R2 and R3 are added to the appendix, and Section 5.2 reports the validation-selected operating point with event sensitivity and delay alongside the test-selected values of Table 3.
[Q3] Table 5 gives CHB-MIT AUC-PR of 0.413 for the full model, 0.389 without retrieval, and 0.406 for the Mamba-only variant. What are the paired fold-level differences? The evidence for persistent state is stronger than the evidence that the explicit-memory construction itself explains the long-sequence gain. Clarifying this distinction would materially improve the contribution claim.
We agree with this reading, and it reflects how the model is designed. The long-sequence gain comes from the recurrent state carried across units. The memory bank is a short-range store, and information older than the bank is held implicitly in the recurrent state, in particular the slow stream.
Table R4 gives the paired fold-level differences: CHB-MIT, 15 channels, NeurDuo-Small, five cross-subject folds, same seed. Each difference is the full model minus the variant on the same test fold.
Table R4 · Paired fold-level AUC-PR differences on CHB-MIT (full model minus variant). The full model reaches an AUC-PR of 0.584 / 0.506 / 0.512 / 0.372 / 0.093 on folds 0–4 (mean 0.413). Each fold entry is the full model minus the variant on that fold, so a positive entry means the full model is higher. The interval is a 95 % paired-t interval over the five folds.
| Group | Variant | Mean AUC-PR | Fold 0 | Fold 1 | Fold 2 | Fold 3 | Fold 4 | Mean difference [95% CI] | t |
|---|---|---|---|---|---|---|---|---|---|
| Persistent state | State reset at every unit (Table 19) | 0.307 | +0.125 | +0.267 | −0.080 | +0.207 | +0.015 | +0.107 [−0.068, +0.282] | +1.69 |
| Persistent state | Shuffled unit order | 0.371 | +0.029 | +0.101 | +0.101 | −0.043 | +0.024 | +0.042 [−0.033, +0.118] | +1.56 |
| Explicit memory | w/o memory retrieval | 0.389 | +0.105 | +0.179 | +0.048 | −0.108 | −0.104 | +0.024 [−0.134, +0.182] | +0.43 |
| Explicit memory | w/o slow stream & memory | 0.394 | +0.167 | −0.114 | +0.067 | −0.048 | +0.024 | +0.019 [−0.115, +0.153] | +0.40 |
| Explicit memory | Mamba-only | 0.406 | +0.099 | +0.094 | −0.087 | −0.069 | −0.002 | +0.007 [−0.102, +0.116] | +0.18 |
Results:
- Persistence is what drives the long-sequence gain.
- Resetting the state at every unit lowers AUC-PR on four of the five folds, by 0.107 on average, and raises false alarms more than fivefold, from 0.96 to 5.15 per hour.
- Shuffling the order of the units keeps every input but breaks their temporal order. It lowers AUC-PR on four of the five folds and raises false alarms to 3.79 per hour.
- The explicit-memory variants stay close to the full model on CHB-MIT. Their differences are small (0.007–0.024) and change sign across folds.
Why this follows from the design.
- The memory bank is a short-range store, matched to the time scale of EEG events.
- It holds eight memory tokens written every 2 s, that is, the last 16 s at 2 s resolution.
- This is where the evidence about an event concentrates. On CHB-MIT, the 185 test seizures last a median of two 30 s units (IQR 2–3), and 88 % last at most four.
- A perturbation analysis confirms that the model draws on past events mostly within this range. It uses NeurDuo-Small models re-trained with our current code, with the same folds and seed.
- In each 10-min test window, we overwrote a 30 s span earlier in the window with a seizure from another recording of the same patient. The scored unit's own input was kept unchanged, and the context head received the original embeddings of all earlier units, so any change in the scored unit's prediction can only come through the backbone's state.
- When the planted seizure fills the 30 s just before the scored unit, it raises the scored unit's seizure probability by 5.07 percentage points.
- Moved 16 s further back, it raises the probability by 0.75 points, 15 % of the immediate effect. Moved 32 s back, by 0.23 points, 5 % of it.
- Most of what the model uses about past events therefore lies in the last 30 s.
- Older information is held implicitly in the recurrent state, in particular the slow stream.
- The planted seizure still raises the prediction when it lies wholly outside the bank: by +0.23 percentage points [+0.05, +0.45] at 32 s, and by +0.07 [+0.01, +0.15] at 60 s (paired 95 % intervals from a subject-level bootstrap).
- Background EEG spliced in the same way has no effect at any lag.
- The slow-stream output keeps a trace of the planted seizure for minutes. After 9 min, it still differs from the original by 0.69 % (median over windows), against 0.09 % for background EEG.
- On CHB-MIT, this time scale is within reach of the recurrent state that every ablation variant keeps. That is why removing retrieval or the slow stream barely moves AUC-PR there, while removing persistence does.
[Q1] How does NeurDuo compare with cached 30-second baseline encoders plus the existing causal context head when measuring the entire inference pipeline at the same output cadence? Please include the task-head parameters and accumulation latency.
We measured this directly. Every model runs its complete CHB-MIT inference pipeline on the same GPU (one NVIDIA GH200, fp32) and emits one decision per 30 s unit, the cadence of Table 3.
Table R1 · Whole-pipeline cost at one decision per 30 s unit, CHB-MIT, 15 channels. Median over the timed repetitions after warm-up; all models on the same GPU.
Results:
At this same cadence, NeurDuo reaches an AUC-PR of 0.392–0.471, against at most 0.208 for the window baselines of Table R1, and it raises the fewest false alarms (Table 3; for CodeBrain, 0.174 AUC-PR and 6.90 false alarms per hour).