- Rating
- 1
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2(+1)
- Soundness
- 1
Read review
Paper summary
The paper proposes StrideDiffusion, a training-free sampler for time series diffusion model. StrideDiffusion tracks band power and changes in power and phase, and uses a spectral gate to decide the stride of timestep to take. The results show that StrideDiffusion outperforms baseline samplers such as DPM-Solver++, UniPC, and DDIM on mean Context-FID metric among most of the datasets.
Critical strengths
The idea of using spectrum information of time series to design the diffusion sampler is interesting.
Critical weaknesses
- The paper is hard to read and understand. First of all, the paper uses lots of notation without explaining what each means and the motivation behind it. Secondly, the methodology is poorly explained. For example, section 3.2 introduces and and uses it as the condition on whether a band is active. However, it is not explained what these two statistics represent, and why we set the condition using the given threshold, and the definition and values of , , and are not explained as well.
- The performance doesn’t seem consistently outperforming baselines in other metrics among most datasets. In the main performance table Table 1 only reports Context-FID as the metric. Other time series generation metrics such as discriminative score and predictive score are listed in Appendix A.1, while the performance are worse than at least one baseline for at least three of the five metrics on datasets Stocks, MuJoCo, ETTh and fMRI.
Questions for authors
Clarification requests:
- What is ? “the estimate at the preceding visited time” means the previous timestep that updates to t? It would be better to define it rigorously.
- Where does the correction equation come from? What is the meaning behind it? Why should we define it this way?
- What do the statistics and mean? Why are they defined like the given equation?
- What are , , ? How do we determine these thresholds?
- Section 3.3 states “Thus the controller uses consecutive visited states“, so it uses the previous one state or all the previous states? If it is all the states, is that used through ?
- Section 3.3 states “The implementation chooses a large stride if is empty, a medium stride if only band 0 (DC) is active, and a small stride otherwise. We set the small stride to one.” What about large stride and medium stride? How large are they?
- Section 4 uses , but in equation 2 is using two timesteps t→s not noise level . Are they the same ? What does mean?
6 comments
Response to Reviewer EApM (part 2 of 3)
Q4. What are , and , and how are they chosen?
is the smallest relative power a band must have to be considered. The relative power of a band is its mean power per frequency, divided by the sum of these means over the four bands. A band below is never active. is the threshold on , so the power of a band counts as changing when its logarithm has moved by at least . is the threshold on . It decreases linearly to half its value by the end of sampling. For the three bands above the lowest one, is multiplied by 1.5 before the comparison. A band is active when its relative power is at least and its power or its phase is changing.
The values are chosen from six candidate settings that were fixed in advance (Appendix A, "Validation candidates"). Each setting gives the three thresholds, the stride sizes and the final window. The factor 1.5 and the decrease of are the same in every setting. For each dataset we ran all six settings on a separate validation model. We kept the one with the lowest validation C-FID among those that use at most 96 network calls. The reported results come from other trained models. Any of the six settings can be selected for any dataset, and Sines and MuJoCo select the same one. Table R4 lists the selected values, which are also in Appendix M.2 of the revision.
Table R4. Settings used for the main results.
| Dataset | Diffusion steps | Large stride | Medium stride | Small stride | Final window | |||
|---|---|---|---|---|---|---|---|---|
| Sines | 500 | 50 | 10 | 1 | 20 | 0.10 | 0.01 | 0.04 |
| Stocks | 500 | 50 | 20 | 1 | 20 | 0.10 | 0.01 | 0.04 |
| ETTh | 500 | 50 | 10 | 1 | 12 | 0.70 | 0.02 | 0.08 |
| MuJoCo | 1000 | 50 | 10 | 1 | 20 | 0.10 | 0.01 | 0.04 |
| Energy | 1000 | 40 | 5 | 1 | 12 | 0.10 | 0.01 | 0.06 |
| fMRI | 1000 | 20 | 5 | 1 | 20 | 0.50 | 0.01 | 0.06 |
Q5. Does the controller use the previous state or all previous states?
Only the previous one, as Section 3.3 of the revision now states. The controller compares the current sample with the sample at the previous visited time. The update uses one stored noise estimate, , from that same time, and nothing older is kept. The word "visited" refers to the times at which the sampler stops, so after a jump the previous visited state is the state before the jump.
Q6. How large are the large and medium strides?
The large stride is 50 diffusion steps on Sines, Stocks, ETTh and MuJoCo, 40 on Energy and 20 on fMRI. The medium stride is 10 on Sines, ETTh and MuJoCo, 20 on Stocks, and 5 on Energy and fMRI. The small stride is always one. For scale, the diffusion model has 500 steps on Sines, Stocks and ETTh and 1000 steps on the other three datasets. These values are in Table R4, and Section 3.3 of the revision now gives them as well.
Q7. Is in Section 4 the same as in Eq. (2)?
Yes, they are the same map, and the revision now says so at the start of Section 4. in Eq. (2) is with and . Both denote one DDIM step from the noise level to the cleaner level , using the clean prediction made at . Here , which grows toward 1 as the sample gets cleaner. Section 4 uses noise levels because its results depend only on the two noise levels. The same formula then also covers the last step to the clean sample, where .
W1. Presentation and notation
The revised Section 3 opens with the plain description given at the start of this reply and with a notation table (Table 1). Each statistic and threshold is now followed by a sentence on what it measures and why it is there.
Response to Reviewer EApM (part 3 of 3)
W2. Results on the other metrics
The main results table of the revision (Table 2) now reports all five metrics, on nine trained models per dataset instead of three. Appendix M.1 gives every sampler on every dataset. Over the five metrics our sampler is best more often than any other sampler. Against each of the other samplers it is significantly better in more cases than it is worse.
The review counts, for each dataset, the metrics on which a sampler is not the best. By this count, every sampler in the main results table falls short on three or more metrics on Stocks, MuJoCo and fMRI. This includes the full DDIM sampler with 500 or 1000 steps. On ETTh the same holds for every sampler except matched uniform replay, which is our own update rule on a uniform grid. The count therefore does not tell the samplers apart. Only twice in the whole table is a sampler best on three or more of the five metrics of a dataset. One of the two cases is our sampler on Energy, with four metrics.
A clearer comparison tests each metric on each dataset, which gives 30 cells. Tables R2 and R3 show the result. Both tables also appear in our comment to all reviewers.
Table R2. All five metrics. Each comparison has 30 cells, one for each dataset and metric. The table counts the cells with a significant difference.
| Compared with | Cells where ours is better | Cells where ours is worse | Cells with no significant difference | Where ours is worse |
|---|---|---|---|---|
| Full DDIM | 15 | 1 | 14 | MuJoCo: HF error |
| UniPC, as in the submission | 16 | 3 | 11 | fMRI: correlation, predictive. MuJoCo: HF error |
| UniPC, grid and order selected on validation | 14 | 3 | 13 | fMRI: correlation, predictive. MuJoCo: HF error |
| Matched uniform replay | 11 | 5 | 14 | ETTh: correlation, predictive. Sines: HF error. MuJoCo: HF error. fMRI: predictive |
| Uniform history, selected budget | 11 | 6 | 13 | ETTh: C-FID, HF error. fMRI: correlation. Sines: HF error. MuJoCo: HF error. Stocks: predictive |
| DPM-Solver++, grid and order selected on validation | 9 | 4 | 17 | fMRI: C-FID, correlation, HF error, predictive |
| DPM-Solver++, as in the submission | 8 | 6 | 16 | fMRI: C-FID, correlation, HF error, predictive. Stocks: HF error. MuJoCo: HF error |
Compared with the full DDIM sampler, ours is better in 15 cells and worse in one, with only 3 to 8 percent of the sampling time.
Table R3. The six samplers of the main results table, ranked by their nine-model mean in each of the 30 cells.
| Sampler | Cells where it is best | Cells among the best two | Cells among the best three |
|---|---|---|---|
| Ours | 10 | 14 | 20 |
| DPM-Solver++ | 6 | 14 | 19 |
| Uniform history, selected budget | 4 | 12 | 15 |
| Matched uniform replay | 4 | 8 | 14 |
| UniPC | 4 | 6 | 10 |
| Full DDIM | 2 | 7 | 13 |
Our sampler has the best mean in 10 of the 30 cells, and the next sampler, DPM-Solver++, in 6. The cells where another sampler is better are listed in the last column of Table R2. Half of them are on one dataset, fMRI. Of the rest, the one that recurs most is the high-frequency error on MuJoCo, and it comes from selecting all settings by validation C-FID alone.
When the setting is selected by the average rank over the five validation metrics, MuJoCo gets a setting with 67 calls instead of 38. Its high-frequency error on the nine models then falls from 0.0077 to 0.0024, the level of the best other samplers (0.0025 for DPM-Solver++). In exchange, C-FID rises from 0.070 to 0.077 and the discriminative score from 0.028 to 0.031. Selection by five metrics picks the setting we already use on Sines, ETTh, Energy and fMRI, and a different setting with about twice the calls on Stocks. Appendix M.6 has the full comparison.
The discriminative score lies between 0.498 and 0.500 for every sampler on Stocks and Energy, so it cannot separate the samplers on these two datasets. Our sampler is best on all four other metrics on Energy. On Stocks it has the best C-FID and the best correlation, which is two of the four metrics that can separate the samplers.
Response to Reviewer EApM (part 1 of 3)
Thank you for the detailed questions. We have revised and extended Section 3, and the answers to all seven questions are now in the paper. Several of the questions concern how the sampler works as a whole, so we start with the method in plain words.
The sampler stops at only some of the diffusion steps, which we call the visited times. At each visited time it first decides how far to jump. It takes the Fourier transform of the current noisy samples and groups the frequencies into four bands. For each band it checks whether the power or the phase has changed since the previous visited time. A band is called active when its relative power is above a threshold and it is still changing. These statistics are averaged over the batch, so all samples in a batch share one grid. If no band is active, the sampler takes a large jump. If only the lowest band is active, it takes a medium jump, and otherwise a single step. Once the time index is at or below the final window K, it always takes single steps.
It then carries out the jump. The update is DDIM with one extra term that uses the noise estimate from the previous visited time. Single steps and the last step to the clean sample use plain DDIM.
Q1. What is ϵ^prev? Is it the estimate at the previous timestep that updates to t?
Yes. It is the noise estimate at the time from which the sampler stepped to t. More precisely, the sampler visits the times t0>t1>t2>…, starting from t0=T−1. When it is at tk, the previous visited time is tk−1, and ϵ^prev is the noise estimate that Eq. (1) gave at tk−1. It was computed when the sampler was there, so it costs no extra network call. After a single step tk−1=tk+1, and after a jump of 50 steps tk−1 lies 50 steps back. At t0 there is no previous estimate, so the sampler uses DDIM. Section 3.1 of the revision gives this definition.
Q2. Where does the correction H come from, and what does it mean?
H is the DDIM step plus one correction term, which comes from the second-order multistep update of the DPM-Solver family (Lu et al., 2022, 2025).
Write αt=ρt, σt=1−ρt, λt=log(αt/σt) and h=λs−λt. The DDIM step of Eq. (2) can then be rewritten exactly as
xs=αtαsxt−σs(eh−1)ϵ^t.
This step keeps the noise estimate constant during the whole jump, which is inaccurate for a long jump because the estimate changes along the way. The multistep update corrects for this. It extends the line through the last two estimates and replaces ϵ^t by
ϵ^t+2hprevh(ϵ^t−ϵ^prev),
where hprev is the value of h for the previous step. Our H is this formula with the constant h0=0.1 in place of hprev.
We use a constant because our grids mix unit steps and long jumps. After a unit step hprev is very small, so the weight h/(2hprev) becomes very large. The update then amplifies the noise in the difference of the two estimates. A constant keeps the weight bounded. The new experiments support this choice (Appendix M.4 of the revision). With the measured hprev and no bound, C-FID gets worse on five of the six datasets. With a bound on hprev, the results are close to those with h0=0.1.
Q3. What do δt and vt mean, and why are they defined this way?
Both compare, one band at a time, the sample at the current visited time t with the sample at the previous visited time u.
δt[B] is the absolute change in the logarithm of the power of band B. It tells us whether the amount of signal in this band has changed since the last visit. The logarithm makes the change relative, so a change of ten percent counts the same in a strong band and in a weak one.
vt[B] measures how far the phases in the band have turned since the last visit. For each frequency we take the phase difference between the two visits and average it over the channels. We then take the root mean square over the frequencies of the band, weighted by their power. The weights keep frequencies with almost no power from dominating. vt[B] tells us whether the oscillations in the band have moved or changed shape even when their power is the same. A sine wave that shifts in time is an example, since its power stays the same while its phase changes.
We use two statistics because a Fourier coefficient has a magnitude and a phase, and δt follows the magnitude while vt follows the phase. When neither has changed by more than its threshold since the last visit, we treat the band as settled and allow a longer jump. A band whose relative power is below τE is left out of this test (see Q4).