- Rating
- 2(+1)
- Confidence
- 3
- Informativeness
- 1
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
Physiological signals are often continuously streamed, with different problems such as missing data. To handle this, the authors propose BioLT. They evaluate BioLT on classification/detection and forecasting across many datasets and achieve strong performances when compared with baselines.
Critical strengths
- they do a lot of experiments
- the experimental setting is well described.
- Lots of details in appendix
Critical weaknesses
- The main baseline comparisons for both detection and forecasting are not too convincing. BioLT is often extremely close to Trans XL, which the authors mention is essentially the same model. This hurts the author's contributions and makes the readers wonder why one would use BioLT in particular.
- I think the paper's scope is kind of all over the place and it seems like many portions should have been expanded upon. For example, the ablation of core components before the conclusion is I think one of the more interesting experiments but the authors provide 2 sentences and the table, without any interpretation of the results.
Overall, I think this is a cool paper, but the authors contributions do not convince me why I would want to use their model over existing ones.
Questions for authors
- Although the authors stress how most baseline models evaluate in 10s windows, if I am not mistaken, all of the inputs are still 10s. I think the memory bank is supposed to give the model access to more "time" then 10s but BioLT is still only evaluated with 10s inputs. Why did the authors not try longer inputs?
- What is BioLT’s performance if the autoregressive stage is removed and the decoder is trained only with the task loss?
- what does 10 binary dataset settings mean in Figure 6? Would you be able to report the per dataset Sens@1FA/h?
- For Figure 8, it seems BioLT actually decreases in performance for Beyond 500 windows where trans-xl does not. However the authors state "In contrast, BioLT achieves the strongest performance both before and after the boundary, with no evident degradation beyond the training length." I am a bit confused by this, and would appreciate any clarifications.
- For Table 4, would the w/o both be equivalent to the trans-xl? If so, for this task, why is the results so much worse than in tables 2 and 3? It sees trans xl is quite competitive with BioLT across all datasets.
5 comments
Response to Reviewer s9Yq (2/3)
W2. The scope is broad; the ablation of core components gets two sentences and no interpretation.
Agreed.
Table 11. Ablation of core components (revised Table 4). is relative to BioLT.
| Configuration | What the model sees | ECG | PPG | ||
|---|---|---|---|---|---|
| BioLT | current chunk + 32-slot memory + | 60.29±0.64 | – | 63.19±0.48 | – |
| w/o | no elapsed-time channel | 58.02±1.71 | −2.27 | 60.61±1.43 | −2.58 |
| w/o mem. bank | current chunk only (≤30 min) | 56.68±0.23 | −3.61 | 57.83±0.34 | −5.36 |
| w/o both | current chunk only, no | 53.82±0.17 | −6.47 | 53.74±0.95 | −9.45 |
The revised ablation gives two clear findings. First, removing cross-chunk memory causes the largest single drop in performance ( AUROC for ECG and for PPG), showing that information beyond the current 30-min chunk contributes to sepsis forecasting. Second, removing the elapsed-time input () reduces AUROC by a further 2.3–2.6 points, showing that treating separated valid windows as if they were uniformly spaced loses useful information. Removing both components produces the largest degradation (/ points), suggesting that long-term context and recording discontinuity provide complementary information.
We will therefore revise Section 5.3 so that the ablation is explicitly linked back to the two problems introduced in Figure 1, infinitely-long context and discontinuity, rather than presenting it as an isolated table.
Overall: why would I use BioLT over existing models?
Thanks. We see three practical gaps in existing approaches that motivate BioLT.
(1) Limited longitudinal context. Snapshot-based biosignal models operate on individual short windows and therefore cannot model how physiological information evolves over longer periods. Most findings in our paper and also in the existing literature have demonstrated the importance of modeling long-term information for biosignals.
(2) Limited handling of recording gaps. Generic sequential models can aggregate temporal context, but they typically do not explicitly represent the irregular interruptions that occur in real biosignal streams. In our ablation, removing the elapsed-time input reduces AUROC by 2.3–2.6 points on MC-MED (Table 11).
(3) Limited streaming beyond a fixed training context. Many sequence models are trained with bounded contexts and are not specifically designed to maintain useful long-term state as the recording continues to grow. In our experiments, several baselines show variable performance beyond the 500-window training context, whereas BioLT maintains continuously updated inference with fixed per-update computation and memory (Table 12).
Table 12. Streaming inference cost of BioLT (App. Table A.17; RTX 4090, batch size 1, median over repeated runs).
| History | Latency | Throughput | Incremental GPU | FLOPs/update |
|---|---|---|---|---|
| 0.5 h (180 windows) | 36.6 ms | 4,913 win/s | 3.47 MB | 386 M |
| 2 h (720 windows) | 37.3 ms | 4,822 win/s | 3.62 MB | 398 M |
| 8 h (2,880 windows) | 36.8 ms | 4,895 win/s | 3.62 MB | 398 M |
| 32 h (11,520 windows) | 37.1 ms | 4,856 win/s | 3.62 MB | 398 M |
BioLT is designed to address these three requirements together, while remaining agnostic to the window-level representation. It can therefore be placed on top of existing domain features or pretrained biosignal foundation models rather than replacing them.
Empirically, across the 21 dataset–feature settings, BioLT is significantly ahead of Transformer-XL and Mamba-3 at 5 of 6 evaluated horizons (Table 10, previous post). Its learned state also retains recoverable long-term information: a linear probe reconstructs the preceding chunk at , compared with 0.86 for the Transformer-XL cache (Table 13).
Table 13. Memory probe, VitalDB ECG AF (), mean over 3 seeds (std ≤0.03 for ); 30–118 positive chunks per distance.
| How far back | 30 min | 60 min | 90 min | 120 min | 180 min |
|---|---|---|---|---|---|
| , mean features of the past chunk | |||||
| BioLT memory | 0.98 | 0.77 | 0.66 | 0.52 | 0.30 |
| Transformer-XL cache | 0.86 | 0.63 | 0.52 | 0.45 | 0.27 |
| Snapshot (current window) | 0.50 | 0.44 | 0.41 | 0.37 | 0.30 |
| AUROC, did an AF event occur in that chunk | |||||
| BioLT memory | 0.68 | 0.68 | 0.68 | 0.66 | 0.70 |
| Transformer-XL cache | 0.60 | 0.50 | 0.56 | 0.58 | 0.48 |
| Snapshot (current window) | 0.52 | 0.49 | 0.48 | 0.60 | 0.48 |
We therefore do not argue that BioLT should replace every existing sequential model. Rather, it is most useful when biosignal monitoring requires the combination of long-term context, explicit handling of interrupted recordings, and continuously updated prediction over an extending stream.
Response to Reviewer s9Yq (3/3)
Q1. All inputs are still 10 s; why not try longer inputs?
(Un)fortunately, we believe this point reflects a misunderstanding, but one that can be resolved quickly by distinguishing token length from context length.
Each token represents one 10 s window, but the context extends over the whole recording. At test time, BioLT streams each recording from its first window to its last, reaching up to 5,907 windows (around 16.4 h) in the original experiments, while carrying the memory throughout.
During fine-tuning, the task loss is applied to segments of at most 500 windows, but the memory entering each segment is initialized from the preceding recording history. The 10 s token length is intentionally retained because BioLT is designed to operate on top of existing window-level feature extractors and foundation models, many of which use fixed short-window inputs.
Therefore, increasing the duration of the individual input window would test a different question: how much local waveform context should be given to the upstream encoder. Our question is instead how these established short-window representations can be modeled over hours-long streaming context. We will make this token/context distinction explicit in Section 3.
Q2. Performance without the autoregressive stage?
Table 14 reports the change in AUROC when AR pretraining is removed: same architecture and fine-tuning, random initialisation, task loss only, against BioLT with pretraining trained in the same code version.
Table 14. Effect of removing AR pretraining: AUROC (points) relative to BioLT, means over 3 seeds.
| Dataset | w/o AR pretraining () | |
|---|---|---|
| CHB-MIT | 0 | −0.8 |
| CHB-MIT | 20 | −2.1 |
| MC-MED ECG | 0 | −1.7 |
| MC-MED ECG | 20 | −3.1 |
| MC-MED PPG | 0 | −0.6 |
| MC-MED PPG | 20 | −0.5 |
| VitalDB ECG | 0 | −0.6 |
| VitalDB ECG | 20 | −1.9 |
Without AR pretraining AUROC drops in all six AF and sepsis settings, by 0.5–3.1 points. It highights the importance of the autoregressive pretraining.
On the other hand, we did experiments to see whether the autoregressive pretraining did actually learn “trajectory dynamics”, rather than just learning to copy paste the last-token values. The results shown below demonstrates the effectiveness of our method compared to other trivial predictors.
Table 15. Trivial-predictor controls for the autoregressive objective. Validation MSE on the target (lower is better); mean over 3 seeds.
| Predictor | CPSC-2021 | MC-MED ECG | VitalDB ECG AF |
|---|---|---|---|
| Trained AR (ours) | |||
| Always-zero (copy previous window) | 0.7979 | 0.7019 | 0.5583 |
| Mean predictor | 0.7979 | 0.7019 | 0.5583 |
| Persistence | 2.3765 | 2.0838 | 1.6396 |
Q3. What does “10 binary dataset settings” mean in Figure 6? Would you report the per-dataset Sens@1FA/h?
Thanks for your question. A setting is one detection task paired with one window representation. Figure 6 averages Sens@1FA/h over these 10:
-
CHB-MIT seizure detection (EEG): domain features, LaBraM, CBraMod (3);
-
VitalDB AF detection, ECG: domain features, ECGFounder, CLEF, CSFM (4);
-
VitalDB AF detection, PPG: domain features, PulsePPG, CSFM (3).
These tasks contain bounded positive episodes with both onset and offset, which allows an event to be classified as detected or missed. MC-MED sepsis is also binary, but its detection label remains positive after onset (sepsis, normally become persistent for a long while after onset) and therefore does not define bounded episodes under our current event-level evaluation. We consequently exclude MC-MED from Sens@1FA/h.
We agree that the original caption was not sufficiently explicit. We will define the 10 settings directly in Figure 6 and add the per-dataset results to the appendix:
Table 16. Sens@1FA/h at detection by dataset (number of feature streams in parentheses).
| Dataset | Snapshot | LSTM | TCN | Mamba-3 | TXL | BioLT |
|---|---|---|---|---|---|---|
| CHB-MIT (3) | 0.111 | 0.018 | 0.063 | 0.181 | 0.251 | 0.253 |
| VitalDB ECG (4) | 0.369 | 0.117 | 0.288 | 0.264 | 0.207 | 0.281 |
| VitalDB PPG (3) | 0.354 | 0.249 | 0.368 | 0.279 | 0.365 | 0.417 |
| Figure 6 average (10) | 0.287 | 0.127 | 0.244 | 0.244 | 0.267 | 0.314 |
The per-dataset table will be added to the appendix.
Q4. Figure 8: BioLT seems to decrease beyond 500 windows while Transformer-XL does not, yet the text says BioLT is strongest before and after.
Thanks for pointing this out. We will correct the wording. Both Transformer-XL and BioLT do not show a significant decline beyond the 500-window boundary, and the observed change remains within the seed variation.
Response to Reviewer s9Yq (1/3)
Thank you for the detailed questions.
Thank you for pointing this out. Our wording “architecturally identical” was confusing, apologies. We meant that Transformer-XL was deliberately matched to BioLT in its basic Transformer backbone, so that it serves as a strong and fair long-context baseline; we did not mean that the complete models are essentially the same. We will revise this wording accordingly.
Table 9. Memory probe, VitalDB ECG AF (H=20), mean over 3 seeds (std ≤0.03 for R2); 30–118 positive chunks per distance.
Transformer-XL is intentionally a stringent comparison in this setting. Despite the close matching, across the 21 dataset–feature settings BioLT is significantly ahead at 5 of 6 evaluated horizons (Table 10), with median AUROC gains of +1.52, +1.76, and +1.53 points at H=10,20,30, respectively. The difference is also visible when we directly probe what the learned state retains: the frozen BioLT memory reconstructs the preceding chunk at R2=0.98, compared with 0.86 for the Transformer-XL cache, with the advantage persisting over longer temporal distances (Table 9).
Especially, our key novelty and also the main contribution is that if we these established sequential modelling (such as Transformer-XL) established in LLMs or somewhere else, what they cannot do? As highlighted in Figure 1 and throughout our paper, there are two main challenges, (1) discontinuity, where valid biosignal windows are separated by irregular gaps that may themselves carry information; and (2) open-ended context, where monitoring streams continue to grow and require efficient retention of long-term physiological history.
More broadly, as also clarified in response to the other reviewers, we do not intend to claim that BioLT is an entirely new architecture for generic sequence modeling, or that its individual building blocks are novel in isolation. Our motivation comes from a different problem: biosignal modeling has largely focused on short “snapshot” windows, while how to extend established window-level representations toward long, irregular, continuously streaming monitoring remains much less clear. Our contribution is therefore in identifying this problem, adapting appropriate and domain-informed approaches to address it, and empirically validating the need for these design choices.
We will therefore revise the paper in two ways: first, to distinguish clearly between a matched Transformer backbone and an identical model; and second, to position the contribution more precisely as a practical streaming-biosignal formulation, rather than novelty of any individual sequence-modeling primitive.
Table 10 is a two-sided paired Wilcoxon test over the 21 dataset–feature settings (three-seed means), as in App. Table A.16. Entry: median AUROC difference (BioLT − baseline) and p; bold: p<0.05.
Table 10. BioLT minus baseline, median over 21 settings (p).
No. “W/o mem. bank” removes all cross-chunk memory, so “w/o both” is a chunked causal decoder that sees only the current chunk and no elapsed time; Transformer-XL keeps a 180-window cache and Δs. Table 4 is also at H=10, Tables 2/3 at H=0 and 20. At H=10 Transformer-XL is close to BioLT, as in Tables 2/3: 60.46±3.38 vs 60.29±0.64 on ECG and 63.91±0.39 vs 63.19±0.48 on PPG, within the seed variation.