- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 3
Read review
Paper summary
This paper tackles the problem of unconditional generation of irregular multivariate time series. The method consists of three conditional flow-matching modules: IntensityFlow, PatternFlow, and ValueFlow, which captures, respectively, observation count and feature-wise frequencies, joint observation time and features, and values conditioned on the generated sampling pattern. The paper also provides evaluation contributions through six complementary diagnostics, real-real references, and controlled corruption experiments. Across five clinical benchmarks, the proposed method achieves the best performance in 21 of 30 dataset-metric comparisons.
Critical strengths
One of the paper’s strengths is a coherent formulation of irregular data generation, an important and under-studied area in generative modeling. The proposed method appears to address a broader and more structurally heterogeneous task than previous proposed methods.
Another strength is the paper’s broad empirical evaluation with multiple metrics. The proposed method achieves the state-of-the-art performance on the majority of experiments across five datasets and evaluation metrics, which includes sampling structure, values, transitions, cross-feature relationships, and an overall discriminator. Corruption tests and oracle interventions also help interpret the metrics and model. Therefore evidence is not confined to a single favorable statistic or dataset.
Critical weaknesses
While the paper offers multiple datasets paired with multiple metrics, the comparisons offered do not necessarily establish advantage over the strongest relevant alternatives. For example, there is no direct comparison with a grid-and-mask generator, despite the central motivation of the paper contrasting native irregular generation with grid-based approaches. The paper also identifies TimEHR but does not evaluate it.
Another weakness of the paper is that there is not enough evidence to clearly isolate the central architecture’s benefits and structural consistency. The proposed hierarchy may work empirically, but the evidence does not establish which components are necessary or how reliably its stages remain mutually consistent. Factorization alternatives differ in capacity and conditioning. There is no isolated ablation of feature-frequency conditioning, nor time-then-panel or panel-then-time comparison. Downstream training uses true conditions, while generation uses model-produced conditions.
Questions for authors
- How does ChronoFlow compare with a strong resolution-matched grid-and-mask generator and a simpler panel-based generation baseline?
- what happens when feature frequency conditioning is removed, and do the principal hierarchy/coupling findings persist beyond eICU?
- how often do sampled frequencies disagree with decoded frequencies, panels require repair, or timestamp clipping create ties?
- What evidence shows preservation of dependencies between observation patterns and values, beyond their separate marginals and local summaries?
7 comments
[W2] & [Q2] Ablation studies of conditioning, Stage-2 ordering, and generalization beyond eICU
1. Which components are necessary, capacity, and mutual consistency. We removed one component at a time on eICU:
| Variant (eICU) | Set-Discr. ↓ | CoOb ↓ | Pattern-MMD² ↓ | Trans-MMD² ↓ | Value-W1 ↓ | Set-Corr. ↓ |
|---|---|---|---|---|---|---|
w/o IntensityFlow (2.18M) | 0.1178 ± 0.0267 | 0.0593 ± 0.0047 | 0.0020 ± 0.0004 | 0.0069 ± 0.0002 | 0.0555 ± 0.0063 | 0.0344 ± 0.0047 |
Joint TPV (size-matched, 2.06M) | 0.3019 ± 0.0508 | 0.0821 ± 0.0032 | 0.0007 ± 0.0000 | 0.0258 ± 0.0012 | 0.2915 ± 0.0049 | 0.0660 ± 0.0010 |
| ChronoFlow (2.06M) | 0.0995 ± 0.0098 | 0.0380 ± 0.0047 | 0.0002 ± 0.0001 | 0.0045 ± 0.0002 | 0.0504 ± 0.0047 | 0.0385 ± 0.0003 |
- Each of these components matters. Removing IntensityFlow degrades the pattern metrics, merging pattern and value generation collapses fidelity, independent per-feature panels break co-observation (
Bernoulli-r, W1: CoOb 0.4383), and removing OT raises Set-Discr. to 0.4316 (Table 2). - Capacity does not explain the gains. Factorizations necessarily differ in conditioning, so we controlled capacity:
Joint TPVfails at matched size, and the largerw/o IntensityFlow(2.18M), Value→Pattern (2.70M: 0.4511), and Joint All (2.21M: 0.2445) are also worse. With the same conditioning (nor), separate pattern and value generation gives Set-Discr. 0.0819 (Monly, point 2) vs. 0.3019 for the size-matchedJoint TPV. - The stages remain mutually consistent. Realized panel frequencies stay within 1.93 percentage points of the sampled ones and follow
ralmost one-to-one (slope 0.9927; Q3), and ChronoFlow preserves the pattern–value coupling closest to Real–Real (D_CV 0.0231 vs. 0.0185; next-best FlexTPP 0.0337; Q4).
2. Feature-frequency conditioning and time–panel order. We removed r (M only) and replaced the joint time–panel PatternFlow with sequential Time-first and Panel-first variants (eICU):
| Variant (eICU) | Params | Set-Discr. ↓ | CoOb ↓ | Pattern-MMD² ↓ | Trans-MMD² ↓ | Value-W1 ↓ | Set-Corr. ↓ |
|---|---|---|---|---|---|---|---|
M only | – | 0.0819 ± 0.0196 | 0.0301 ± 0.0084 | 0.0002 ± 0.0001 | 0.0049 ± 0.0004 | 0.0487 ± 0.0029 | 0.0399 ± 0.0042 |
Time-first | 2.13M | 0.0751 ± 0.0136 | 0.0324 ± 0.0069 | 0.0004 ± 0.0001 | 0.0039 ± 0.0003 | 0.0537 ± 0.0074 | 0.0388 ± 0.0032 |
Panel-first | 2.14M | 0.0733 ± 0.0256 | 0.0322 ± 0.0060 | 0.0002 ± 0.0000 | 0.0046 ± 0.0001 | 0.0523 ± 0.0012 | 0.0326 ± 0.0043 |
| ChronoFlow (joint) | 2.06M | 0.0995 ± 0.0098 | 0.0380 ± 0.0047 | 0.0002 ± 0.0001 | 0.0045 ± 0.0002 | 0.0504 ± 0.0047 | 0.0385 ± 0.0003 |
ris designed for controllability. Removing it does not consistently change unconditional fidelity (better on some metrics, worse on others), since PatternFlow learns marginal rates implicitly. However,rlets a practitioner specify how often each feature is observed, which is impossible withMonly: withMand the noise fixed, changingrmoves the realized frequencies almost one-to-one (slope 0.9927).- The time–panel order is flexible. The sequential variants reach similar fidelity, so the hierarchy does not hinge on this order; what matters is generating each occasion's features together and the pattern before the values (point 1). The joint form needs a single flow, whereas the sequential variants split Stage 2 in two, with 3.3–3.8% more parameters and 7–8% longer training. We have revised Section 5.3 and added this comparison to Appendix G.1.
[W2] & [Q2], continued
3. Training on true conditions vs. sampling on generated ones. Conditioning crosses only two coarse-to-fine boundaries, not hundreds of autoregressive steps in which errors compound. Replacing generated conditions with ground-truth ones, and perturbing them:
| Setting (eICU, Table A4) | (M, r) | (τ, B) | Value-W1 ↓ | Set-Corr. ↓ |
|---|---|---|---|---|
| Full generation | sampled | sampled | 0.0504 ± 0.0047 | 0.0385 ± 0.0003 |
+ ground-truth (M, r) | ground truth | sampled | 0.0481 ± 0.0037 | 0.0365 ± 0.0022 |
+ ground-truth (M, r), (τ, B) | ground truth | ground truth | 0.0508 ± 0.0100 | 0.0358 ± 0.0019 |
| Ground truth + 10% noise | perturbed | perturbed | 0.0460 ± 0.0015 | 0.0378 ± 0.0049 |
| Ground truth + 20% noise | perturbed | perturbed | 0.0473 ± 0.0026 | 0.0364 ± 0.0050 |
Ground-truth patterns leave the value marginals essentially unchanged and values stay stable under 10–20% noise, while ground-truth (M, r) lowers Set-Discr. by 85% (0.0995 → 0.0153), so Stage 1 is the main source of the Set-Discr. gap. We have added mitigation paths (a better Stage-1 sampler, conditioning-noise augmentation, and self-distillation) to Section 5.4.
Q2. Removing feature-frequency conditioning, and findings beyond eICU. Removing r is answered in point 2. On the non-clinical Activity dataset, retraining PatternFlow without OT (w/o OT) is worse than ChronoFlow on all six metrics and merging all stages (Joint All) on five of six, so the hierarchy and OT findings persist beyond clinical EHR:
| Variant (Activity) | Set-Discr. ↓ | CoOb ↓ | Pattern-MMD² ↓ | Trans-MMD² ↓ | Value-W1 ↓ | Set-Corr. ↓ |
|---|---|---|---|---|---|---|
| ChronoFlow | 0.1753 ± 0.0217 | 0.0739 ± 0.0100 | 0.0278 ± 0.0080 | 0.0100 ± 0.0015 | 0.1018 ± 0.0068 | 0.0474 ± 0.0067 |
w/o OT | 0.2292 ± 0.0376 | 0.1360 ± 0.0053 | 0.0355 ± 0.0031 | 0.0195 ± 0.0016 | 0.1067 ± 0.0062 | 0.0512 ± 0.0069 |
Joint All | 0.2691 ± 0.0552 | 0.1589 ± 0.0055 | 0.0200 ± 0.0022 | 0.0431 ± 0.0195 | 0.1438 ± 0.0099 | 0.0625 ± 0.0093 |
We thank the reviewer for the precise questions and for recognizing (1) our coherent formulation of irregular time-series generation, an important, under-studied, and structurally heterogeneous problem, and (2) our broad evaluation, in which ChronoFlow leads on most dataset–metric pairs and corruption tests and oracle interventions keep the evidence from resting on a single favorable statistic. We address each concern below.
[W1] & [Q1] Comparison with resolution-matched grid-and-mask generators and a simpler panel-based baseline
We compared ChronoFlow on eICU, with the same event representation and six metrics, against two resolution-matched grid-and-mask generators, each given a fair chance, and a simpler panel-based baseline:
Bernoulli-r: IntensityFlow, the generated times, and ValueFlow are kept, but each occasion's panel is drawn as independent Bernoulli(r_f) per feature, preserving the sampled rates but ignoring co-observation.Bernoulli-rBernoulli-rmatches the sampled rates yet is almost perfectly separable and breaks co-observation, so modeling which features are measured together, PatternFlow's role, is essential.We have added
Bernoulli-rto Table 2 and Section 5.3, and we will add TimEHR and TimeDiff to Table 1.[1] Karami et al. "TimEHR: Image-Based Time Series Generation for Electronic Health Records." IEEE Journal of Biomedical and Health Informatics 2025.
[2] Tian et al. "Reliable Generation of Privacy-Preserving Synthetic Electronic Health Record Time Series via Diffusion Models." JAMIA 2024.