- Rating
- 2
- Confidence
- 3
- Informativeness
- 1
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This work proposes PRISM, an MLP-based model that recursively branches input segments and adaptively selects forecast-relevant information from different temporal segments using Haar wavelet transforms. The method is evaluated on the LTSF and GIFT-Eval benchmarks against several prominent forecasting models and achieves strong results.
Critical strengths
-
The use of wavelets across both global and predominantly local segments is interesting.
-
While many recent methods combine spectral transformations with patches and Transformers, PRISM explores a relatively simple MLP-based alternative, which is an interestingly simple and practical direction.
-
The paper provides fair comparisons by using the best results reported in the original papers or through reproduction. This is particularly valuable in LTSF, where many works rely on their own reproduced baselines.
-
The authors evaluate their results on two established benchmarks, thus adding to the robustness of their approach.
Critical weaknesses
1. Novelty
Spectral and wavelet decomposition have been widely explored in deep-learning-based TSF. While PRISM achieves strong results, many of its components build on established techniques, with the main contribution appearing to lie in architectural design.
In particular, WPMixer [1], WaveletMixer [2], and WDformer [3] also employ localized/patch-based wavelet decomposition. The paper should more clearly distinguish PRISM from these closely related approaches and articulate its novel contribution. More importantly, compare their performance.
[1] WPMixer: Efficient Multi-Resolution Mixing for Long-Term Time Series Forecasting (AAAI 2025)
[2] WaveletMixer: A Multi-Resolution Wavelets Based MLP-Mixer for Multivariate Long-Term Time Series Forecasting (AAAI 2025)
[3] WDformer: A Wavelet-based Differential Transformer Model for Time Series Forecasting (CIKM 2025)
2. Experiments and Results
-
Why are SCINet and WPMixer omitted from Table 1? They are discussed and included in the significance analysis, but no explanation is provided for their absence from the main comparison.
-
The baselines appear to be relatively outdated, with methods introduced in 2024 or earlier. Given the rapid development of TSF, the evaluation would benefit from including more recent SOTA methods, particularly recent approaches using decomposition or spectral components.
-
The inference analysis is limited and relegated to the appendix. I suggest including additional baselines such as iTransformer, PatchTST, and relevant wavelet-based methods. Since wavelet transforms may introduce computational overhead, a more comprehensive efficiency analysis would strengthen the paper and could be moved to the main text.
3. Clarity and Presentation
- In Table 1, the Exchange 336 MAE result for PRISM appears to be missing bold formatting.
- Some appendix/main text tables appear incomplete or inconsistent; please check the “Best count” rows in Tables 2,12, 13, etc.
- Line 201 states that ties are credited to every tied method, potentially making the win counts exceed 32. Please clarify this.
- Section 4.2 feels somewhat unfocused and requires the reader to jump between multiple tables. A more direct discussion of the main findings would make the analysis substantially clearer.
Questions for authors
see above
4 comments
(W2) 2.2 More recent baselines
The baselines are from 2024 or earlier. Recent methods, in particular decomposition- or spectrum-based ones, should be included.
We also provide the comparisons with more recent baselines. Across these, PRISM shows the best MSE results in 27 of 32 settings and the best MAE results in 32 out of 32 settings. These additional comparisons to newer works further strengthen the empirical contributions of our work.
| Dataset | H | PRISM MSE | PRISM MAE | FilterNet MSE | FilterNet MAE | FreqCycle MSE | FreqCycle MAE | AWEMixer MSE | AWEMixer MAE | WDformer MSE | WDformer MAE |
|---|---|---|---|---|---|---|---|---|---|---|---|
| ETTh1 | 96 | 0.355 | 0.374 | 0.424 | 0.426 | 0.371 | 0.397 | 0.371 | 0.397 | 0.415 | 0.424 |
| ETTh1 | 192 | 0.390 | 0.405 | 0.453 | 0.449 | 0.402 | 0.418 | 0.406 | 0.416 | 0.463 | 0.459 |
| ETTh1 | 336 | 0.386 | 0.407 | 0.435 | 0.435 | 0.426 | 0.436 | 0.433 | 0.433 | 0.469 | 0.464 |
| ETTh1 | 720 | 0.421 | 0.445 | 0.450 | 0.461 | 0.454 | 0.475 | 0.466 | 0.473 | 0.650 | 0.577 |
| ETTh2 | 96 | 0.261 | 0.322 | 0.344 | 0.388 | 0.285 | 0.347 | 0.287 | 0.347 | 0.338 | 0.378 |
| ETTh2 | 192 | 0.311 | 0.359 | 0.359 | 0.397 | 0.346 | 0.388 | 0.347 | 0.387 | 0.425 | 0.425 |
| ETTh2 | 336 | 0.308 | 0.361 | 0.386 | 0.423 | 0.374 | 0.414 | 0.380 | 0.414 | 0.469 | 0.450 |
| ETTh2 | 720 | 0.387 | 0.420 | 0.411 | 0.441 | 0.425 | 0.450 | 0.418 | 0.449 | 0.468 | 0.465 |
| ETTm1 | 96 | 0.282 | 0.326 | 0.314 | 0.356 | 0.288 | 0.342 | 0.291 | 0.342 | 0.316 | 0.367 |
| ETTm1 | 192 | 0.323 | 0.352 | 0.340 | 0.378 | 0.328 | 0.366 | 0.331 | 0.367 | 0.355 | 0.390 |
| ETTm1 | 336 | 0.354 | 0.377 | 0.384 | 0.403 | 0.360 | 0.385 | 0.368 | 0.389 | 0.385 | 0.409 |
| ETTm1 | 720 | 0.423 | 0.409 | 0.435 | 0.432 | 0.420 | 0.418 | 0.423 | 0.420 | 0.445 | 0.441 |
| ETTm2 | 96 | 0.158 | 0.234 | 0.179 | 0.262 | 0.164 | 0.253 | 0.174 | 0.257 | 0.194 | 0.276 |
| ETTm2 | 192 | 0.214 | 0.267 | 0.243 | 0.306 | 0.222 | 0.292 | 0.235 | 0.302 | 0.272 | 0.324 |
| ETTm2 | 336 | 0.264 | 0.319 | 0.289 | 0.336 | 0.275 | 0.330 | 0.279 | 0.332 | 0.324 | 0.357 |
| ETTm2 | 720 | 0.352 | 0.372 | 0.373 | 0.394 | 0.362 | 0.385 | 0.365 | 0.387 | 0.405 | 0.409 |
| Traffic | 96 | 0.362 | 0.237 | 0.381 | 0.268 | 0.394 | 0.268 | 0.394 | 0.280 | 0.353 | 0.252 |
| Traffic | 192 | 0.376 | 0.244 | 0.401 | 0.277 | 0.411 | 0.276 | 0.409 | 0.281 | 0.367 | 0.262 |
| Traffic | 336 | 0.401 | 0.257 | 0.413 | 0.282 | 0.423 | 0.281 | 0.433 | 0.294 | 0.376 | 0.265 |
| Traffic | 720 | 0.433 | 0.271 | 0.440 | 0.294 | 0.451 | 0.298 | 0.463 | 0.306 | 0.402 | 0.276 |
| Electricity | 96 | 0.122 | 0.219 | 0.133 | 0.227 | 0.129 | 0.224 | 0.133 | 0.228 | 0.131 | 0.226 |
| Electricity | 192 | 0.145 | 0.235 | 0.151 | 0.244 | 0.146 | 0.240 | 0.150 | 0.243 | 0.154 | 0.248 |
| Electricity | 336 | 0.159 | 0.243 | 0.167 | 0.261 | 0.163 | 0.258 | 0.168 | 0.261 | 0.168 | 0.263 |
| Electricity | 720 | 0.194 | 0.281 | 0.205 | 0.294 | 0.201 | 0.292 | 0.206 | 0.293 | 0.198 | 0.293 |
| Exchange | 96 | 0.081 | 0.198 | 0.098 | 0.223 | 0.093 | 0.216 | 0.098 | 0.221 | 0.113 | 0.240 |
| Exchange | 192 | 0.173 | 0.297 | 0.222 | 0.340 | 0.194 | 0.314 | 0.190 | 0.311 | 0.210 | 0.334 |
| Exchange | 336 | 0.314 | 0.396 | 0.365 | 0.440 | 0.359 | 0.433 | 0.377 | 0.448 | 0.415 | 0.476 |
| Exchange | 720 | 0.712 | 0.606 | 0.928 | 0.724 | 0.942 | 0.729 | 0.975 | 0.746 | 0.942 | 0.738 |
| Weather | 96 | 0.140 | 0.177 | 0.148 | 0.200 | 0.149 | 0.202 | 0.148 | 0.199 | 0.166 | 0.216 |
| Weather | 192 | 0.187 | 0.227 | 0.189 | 0.239 | 0.192 | 0.245 | 0.189 | 0.239 | 0.210 | 0.257 |
| Weather | 336 | 0.234 | 0.266 | 0.242 | 0.280 | 0.246 | 0.286 | 0.240 | 0.280 | 0.263 | 0.295 |
| Weather | 720 | 0.310 | 0.320 | 0.323 | 0.335 | 0.330 | 0.341 | 0.315 | 0.332 | 0.334 | 0.342 |
| Best count | 27 | 32 | 0 | 0 | 1 | 0 | 0 | 0 | 4 | 0 |
(W2) 2.3 Efficiency analysis
The inference analysis is limited and in the appendix. Please add iTransformer, PatchTST and wavelet-based methods, quantify the overhead of the wavelet transform, and consider moving the analysis to the main text.
We thank the Reviewer for this suggestion and have extended the efficiency analysis to the suggested baselines. In addition to D-PAD, DLinear, TimeMixer and XPatch, the revised table includes iTransformer, PatchTST and the two wavelet-based methods WPMixer and WDformer. All models are built from their released code with their ETTh1 hyperparameters and measured under one setting (context 336, horizon 96, batch 8, fp32, one GPU). We report parameter count, FLOPs per sample, which depends only on the model and not on the hardware, and peak memory at inference and during training. The table lists every model relative to PRISM.
| Model | Params (M) | MFLOPs / sample | Inference peak mem (MB) | Train peak mem (MB) |
|---|---|---|---|---|
| PRISM | 0.35 | 5.1 | 13.7 | 26.6 |
| DLinear | 0.10 (0.3×) | 0.9 (0.2×) | 9.0 (0.7×) | 17.9 (0.7×) |
| iTransformer | 0.25 (0.7×) | 5.7 (1.1×) | 10.7 (0.8×) | 21.7 (0.8×) |
| XPatch | 0.33 (0.9×) | 9.8 (1.9×) | 14.5 (1.1×) | 33.9 (1.3×) |
| WDformer | 2.15 (6.1×) | 41.9 (8.2×) | 18.8 (1.4×) | 53.5 (2.0×) |
| D-PAD | 3.22 (9.2×) | 46.8 (9.2×) | 25.0 (1.8×) | 78.2 (2.9×) |
| TimeMixer | 0.73 (2.1×) | 182.6 (36×) | 27.8 (2.0×) | 83.0 (3.1×) |
| WPMixer | 2.26 (6.4×) | 538.3 (106×) | 29.0 (2.1×) | 145.7 (5.5×) |
| PatchTST | 5.23 (14.9×) | 1909.2 (374×) | 97.8 (7.1×) | 238.5 (9.0×) |
Compute. Comparison shows that PRISM is among the lightest models of the table. It needs 5.1 MFLOPs per sample, the same order as DLinear and iTransformer, and 8 to 374 times fewer than the other decomposition-based models, WDformer (41.9 M), D-PAD (46.8 M), TimeMixer (183 M), WPMixer (538 M) and PatchTST (1909 M). With 0.35 M parameters it is smaller than every model except DLinear and iTransformer, and its training memory of 27 MB is a third of D-PAD and TimeMixer and a fifth of WPMixer.
Wavelet overhead. The Haar decomposition is a fixed linear operator with 0.18 MFLOPs per sample in this setting, 3.5% of PRISM's total of 5.1 MFLOPs, and it runs once per node of the temporal tree. The two wavelet baselines cost 8 and 100 times more per sample than PRISM in total, so the overhead of these methods comes from what is built on top of the transform, not from the transform itself. We move the table to Section 4.2.
(W3) 3.1 Missing bold in Table 1
The Exchange-336 MAE entry of PRISM lacks bold.
Thank you for highlighting this. We have fixed the missing bold and the Best-count rows.
(W3) 3.2 Best-count rows and tie counting
Some Best-count rows (Tables 2, 12, 13) look inconsistent. Line 201 says ties are credited to every tied method, so counts can exceed 32.
We would like to clarify that, as a setting, we consider a (dataset, horizon) pair, and a method is credited whenever its value equals the row minimum at the reported precision. When two methods tie both are credited, so column sums can exceed 32. Throughout the text, we used the same approach as specified in Line 201. To address the reviewer’s comment, we added clarifications regarding best counts throughout the tables.
(W3) 3.3 Section 4.2
Section 4.2 is unfocused and jumps between tables.
We thank the Reviewer for highlighting this, as Section 4.2 is the main results section and has to be clear and easy to follow. To address the Reviewer’s comment, we rewrote Section 4.2 so that each paragraph makes one claim and points to one table. The revised section now reads in this order:
-
Setup: datasets, context length, horizons and baselines.
-
Main result: Table 1 and the best-count rule.
-
Significance: paired t-tests with FDR correction (Tables A.4 and A.5).
-
Robustness: the same comparison at context 96 (Table A.14).
Hyperparameter protocols are referenced once at the end (Tables A.15 to A.17). The full text is in the revised PDF.
I thank the authors for the detailed response and the additional experiments.
The new comparisons with wavelet-based and recent baselines and the clarification of the best-count rule address several of my concerns. However, some concerns remain
-
Novelty. I still have concerns about the novelty of PRISM compared with WPMixer, WaveletMixer and WDformer.
-
Efficiency numbers. The new numbers do not match the paper. For ETTh1 (context 336, horizon 96), PRISM is reported at 1.68M FLOPs (Table A.1), 2.52M (Table A.18) and 5.1M (rebuttal). DLinear goes from 679K parameters and 3.63M FLOPs to 0.10M and 0.9M, which reverses the paper’s claim that PRISM needs fewer FLOPs than DLinear. Peak memory goes from 235 MB to 13.7 MB. Latency is also missing from the new table, although Table A.1 shows PRISM to be slower than DLinear and TimeMixer.
-
Hyperparameter tuning and fairness of the new comparisons. PRISM’s hyperparameters appear to have been tuned on the test set. The search in Tables A.11–A.13 reports test results identical to Table 1, and the selected values are the ones with the best test scores. The released scripts also set hyperparameters per dataset and horizon, and many of these settings are missing from Table A.15.
Therefore, based on the remaining concerns about novelty, the inconsistent efficiency results, and the evaluation protocol, I keep my score.
Response to Reviewer chNb
We thank the Reviewer chNb for providing the feedback on our paper. Following your suggestions, we added and compared more recent decomposition-based baselines, extended the efficiency analysis to iTransformer, PatchTST and the wavelet-based methods, and revised the presentation of Section 4.2. We hope what we have done here conveys our effort to address your concerns.
(W1) Novelty relative to WPMixer, WaveletMixer and WDformer
Many components build on established techniques and the contribution appears to be architectural. WPMixer, WaveletMixer and WDformer also use localized or patch-based wavelet decomposition. Please distinguish PRISM from them and, more importantly, compare their performance.
To follow the Reviewer’s suggestion, we compared PRISM’s performance with that of the suggested related works. We observed that, across the standard evaluation settings (combined across the forecasting horizons for brevity), PRISM shows the best MSE performance on 7 out of 8 settings and the best MAE performance on 8 out of 8 settings. These additional comparisons further strengthen the empirical results of our work.
Higher performance of our model comes with architectural differences. WPMixer, WaveletMixer and WDformer all have rigid rules for combining components of decomposed time-series signals, where the mixing is learned once and applied to every input. PRISM recomputes the allocation per sample and lets the same band be treated differently in different parts of the context window. The ablations in Table 6 and Table A.2 show that this design contributes to the increased performance. Replacing tree-structured selection with flat multiscale aggregation worsens the MSE by 5.79%, and replacing explicit selection with implicit mixing worsens it by 4.18%. Both ablations use the same decomposition of the time-series signal, so the contribution of the selection mechanism is separate from that of the decomposition, and separate from what’s done in prior works.
(W2) 2.1 SCINet and WPMixer in Table 1
Why are SCINet and WPMixer omitted from Table 1 although they appear in the significance analysis?
To address the Reviewer’s comment, we provide the requested values below. Across these settings, PRISM shows the best MSE results in 8 out of 8 settings (combined over forecasting horizons here for brevity) and the best MAE in 8 out of 8 settings as well. We did not include these results in Table 1 because, as the counts of relevant baselines have been growing, we aimed to keep the font size of our main results legible.