- Rating
- 4
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper addresses incomplete multi-view clustering by proposing SPORT, which aims to preserve cross-view shared information while retaining view-specific characteristics. Its main components include a structure-aware relational contrastive objective, a prototype disentanglement mechanism based on shared and view-specific subspaces, and a hybrid imputation strategy that combines prototype matching with local neighborhood information. Experiments on six datasets at different missing rates, together with baseline comparisons and ablation studies, are presented to demonstrate the effectiveness of the proposed framework.
Critical strengths
Rather than enforcing complete prototype alignment across views, SPORT explicitly separates shared and view-specific components. This is a meaningful design choice for heterogeneous multi-view data. Extensive experiments further demonstrate the effectiveness of the proposed method.
Critical weaknesses
1.Prototype-based estimates and neighborhood-based estimates may have different reliability depending on the sample and the observed views. Consequently, does the use of fixed mixing coefficients limit the adaptability of the imputation process?
2.The relational weights are derived from cross-view similarity information. When only a small number of views are jointly observed, these similarity estimates will become less reliable, potentially affecting the contrastive learning process.
3.How sensitive is SPORT to the mixing coefficient used to combine prototype and neighborhood information? Could the authors report results for a few representative coefficient values on one or two datasets?
4.When the number of jointly observed views is small, how does the method avoid unreliable similarity estimates in the relational contrastive objective?
5.In the convergence analysis, why is it not the case that the changes in the loss value are shown, but only the changes in performance are presented?
Questions for authors
See in Critical Weaknesses.
4 comments
Reply 2
Table 1: Training loss on Digit4k during pretraining with 0.5 missing rate.
| Epoch | Loss | Epoch | Loss | Epoch | Loss | Epoch | Loss | Epoch | Loss |
|---|---|---|---|---|---|---|---|---|---|
| 2 | 0.103 | 42 | 0.033 | 82 | 0.012 | 122 | 0.006 | 162 | 0.004 |
| 4 | 0.049 | 44 | 0.032 | 84 | 0.012 | 124 | 0.006 | 164 | 0.004 |
| 6 | 0.041 | 46 | 0.031 | 86 | 0.011 | 126 | 0.006 | 166 | 0.004 |
| 8 | 0.039 | 48 | 0.029 | 88 | 0.011 | 128 | 0.006 | 168 | 0.004 |
| 10 | 0.038 | 50 | 0.027 | 90 | 0.011 | 130 | 0.006 | 170 | 0.004 |
| 12 | 0.038 | 52 | 0.026 | 92 | 0.010 | 132 | 0.006 | 172 | 0.004 |
| 14 | 0.038 | 54 | 0.024 | 94 | 0.010 | 134 | 0.005 | 174 | 0.004 |
| 16 | 0.038 | 56 | 0.022 | 96 | 0.010 | 136 | 0.005 | 176 | 0.003 |
| 18 | 0.038 | 58 | 0.021 | 98 | 0.009 | 138 | 0.005 | 178 | 0.003 |
| 20 | 0.038 | 60 | 0.019 | 100 | 0.009 | 140 | 0.005 | 180 | 0.003 |
| 22 | 0.038 | 62 | 0.018 | 102 | 0.009 | 142 | 0.005 | 182 | 0.003 |
| 24 | 0.038 | 64 | 0.017 | 104 | 0.008 | 144 | 0.005 | 184 | 0.003 |
| 26 | 0.038 | 66 | 0.016 | 106 | 0.008 | 146 | 0.005 | 186 | 0.003 |
| 28 | 0.038 | 68 | 0.016 | 108 | 0.008 | 148 | 0.005 | 188 | 0.003 |
| 30 | 0.038 | 70 | 0.015 | 110 | 0.008 | 150 | 0.005 | 190 | 0.003 |
| 32 | 0.037 | 72 | 0.014 | 112 | 0.007 | 152 | 0.004 | 192 | 0.003 |
| 34 | 0.036 | 74 | 0.014 | 114 | 0.007 | 154 | 0.004 | 194 | 0.003 |
| 36 | 0.036 | 76 | 0.013 | 116 | 0.007 | 156 | 0.004 | 196 | 0.003 |
| 38 | 0.035 | 78 | 0.013 | 118 | 0.007 | 158 | 0.004 | 198 | 0.003 |
| 40 | 0.034 | 80 | 0.012 | 120 | 0.007 | 160 | 0.004 | 200 | 0.003 |
Table 2: Training loss on Digit4k during fine-tuning with 0.5 missing rate.
| Epoch | Loss | Epoch | Loss | Epoch | Loss | Epoch | Loss | Epoch | Loss |
|---|---|---|---|---|---|---|---|---|---|
| 202 | 591.593 | 262 | 283.146 | 322 | 204.786 | 382 | 203.208 | 442 | 202.876 |
| 204 | 588.490 | 264 | 258.655 | 324 | 204.707 | 384 | 203.176 | 444 | 202.876 |
| 206 | 585.559 | 266 | 243.328 | 326 | 204.632 | 386 | 203.143 | 446 | 202.881 |
| 208 | 583.321 | 268 | 232.693 | 328 | 204.561 | 388 | 203.111 | 448 | 202.880 |
| 210 | 581.521 | 270 | 225.369 | 330 | 204.492 | 390 | 203.079 | 450 | 202.880 |
| 212 | 579.750 | 272 | 220.213 | 332 | 204.426 | 392 | 203.047 | 452 | 202.879 |
| 214 | 578.309 | 274 | 216.583 | 334 | 204.362 | 394 | 203.013 | 454 | 202.880 |
| 216 | 576.831 | 276 | 213.988 | 336 | 204.301 | 396 | 202.979 | 456 | 202.880 |
| 218 | 575.392 | 278 | 212.109 | 338 | 204.241 | 398 | 202.950 | 458 | 202.879 |
| 220 | 573.933 | 280 | 210.727 | 340 | 204.183 | 400 | 202.930 | 460 | 202.881 |
| 222 | 572.445 | 282 | 209.694 | 342 | 204.126 | 402 | 202.910 | 462 | 202.881 |
| 224 | 570.899 | 284 | 208.907 | 344 | 204.071 | 404 | 202.892 | 464 | 202.881 |
| 226 | 569.276 | 286 | 208.296 | 346 | 204.016 | 406 | 202.880 | 466 | 202.882 |
| 228 | 567.570 | 288 | 207.810 | 348 | 203.962 | 408 | 202.878 | 468 | 202.880 |
| 230 | 565.709 | 290 | 207.415 | 350 | 203.909 | 410 | 202.882 | 470 | 202.880 |
| 232 | 563.672 | 292 | 207.087 | 352 | 203.856 | 412 | 202.881 | 472 | 202.880 |
| 234 | 561.445 | 294 | 206.808 | 354 | 203.803 | 414 | 202.880 | 474 | 202.880 |
| 236 | 558.944 | 296 | 206.565 | 356 | 203.751 | 416 | 202.877 | 476 | 202.879 |
| 238 | 556.069 | 298 | 206.350 | 358 | 203.698 | 418 | 202.878 | 478 | 202.880 |
| 240 | 552.676 | 300 | 206.158 | 360 | 203.647 | 420 | 202.880 | 480 | 202.879 |
| 242 | 548.562 | 302 | 205.982 | 362 | 203.594 | 422 | 202.877 | 482 | 202.879 |
| 244 | 543.327 | 304 | 205.820 | 364 | 203.542 | 424 | 202.877 | 484 | 202.879 |
| 246 | 536.343 | 306 | 205.670 | 366 | 203.490 | 426 | 202.875 | 486 | 202.879 |
| 248 | 526.511 | 308 | 205.530 | 368 | 203.442 | 428 | 202.876 | 488 | 202.878 |
| 250 | 511.906 | 310 | 205.399 | 370 | 203.407 | 430 | 202.876 | 490 | 202.879 |
| 252 | 489.314 | 312 | 205.274 | 372 | 203.373 | 432 | 202.875 | 492 | 202.879 |
| 254 | 455.026 | 314 | 205.157 | 374 | 203.338 | 434 | 202.876 | 494 | 202.878 |
| 256 | 410.483 | 316 | 205.052 | 376 | 203.305 | 436 | 202.877 | 496 | 202.880 |
| 258 | 366.153 | 318 | 204.958 | 378 | 203.272 | 438 | 202.876 | 498 | 202.879 |
| 260 | 323.197 | 320 | 204.869 | 380 | 203.241 | 440 | 202.876 | 500 | 202.878 |
Reply 3
Table 3: Training loss on 100Leaves during pretraining with 0.5 missing rate.
| Epoch | Loss | Epoch | Loss | Epoch | Loss | Epoch | Loss | Epoch | Loss |
|---|---|---|---|---|---|---|---|---|---|
| 2 | 0.122 | 42 | 0.014 | 82 | 0.007 | 122 | 0.005 | 162 | 0.004 |
| 4 | 0.052 | 44 | 0.013 | 84 | 0.007 | 124 | 0.005 | 164 | 0.004 |
| 6 | 0.052 | 46 | 0.013 | 86 | 0.007 | 126 | 0.005 | 166 | 0.004 |
| 8 | 0.048 | 48 | 0.012 | 88 | 0.007 | 128 | 0.005 | 168 | 0.004 |
| 10 | 0.045 | 50 | 0.012 | 90 | 0.007 | 130 | 0.005 | 170 | 0.004 |
| 12 | 0.043 | 52 | 0.012 | 92 | 0.007 | 132 | 0.005 | 172 | 0.004 |
| 14 | 0.037 | 54 | 0.011 | 94 | 0.006 | 134 | 0.005 | 174 | 0.004 |
| 16 | 0.035 | 56 | 0.011 | 96 | 0.006 | 136 | 0.005 | 176 | 0.004 |
| 18 | 0.032 | 58 | 0.011 | 98 | 0.006 | 138 | 0.005 | 178 | 0.004 |
| 20 | 0.030 | 60 | 0.010 | 100 | 0.006 | 140 | 0.005 | 180 | 0.004 |
| 22 | 0.029 | 62 | 0.010 | 102 | 0.006 | 142 | 0.005 | 182 | 0.004 |
| 24 | 0.027 | 64 | 0.009 | 104 | 0.006 | 144 | 0.005 | 184 | 0.004 |
| 26 | 0.025 | 66 | 0.009 | 106 | 0.006 | 146 | 0.005 | 186 | 0.004 |
| 28 | 0.023 | 68 | 0.009 | 108 | 0.006 | 148 | 0.005 | 188 | 0.004 |
| 30 | 0.021 | 70 | 0.009 | 110 | 0.006 | 150 | 0.005 | 190 | 0.004 |
| 32 | 0.019 | 72 | 0.008 | 112 | 0.006 | 152 | 0.005 | 192 | 0.004 |
| 34 | 0.018 | 74 | 0.008 | 114 | 0.006 | 154 | 0.005 | 194 | 0.004 |
| 36 | 0.017 | 76 | 0.008 | 116 | 0.006 | 156 | 0.005 | 196 | 0.004 |
| 38 | 0.016 | 78 | 0.007 | 118 | 0.005 | 158 | 0.004 | 198 | 0.004 |
| 40 | 0.015 | 80 | 0.007 | 120 | 0.005 | 160 | 0.004 | 200 | 0.004 |
Table 4: Training loss on 100Leaves during fine-tuning with 0.5 missing rate.
| Epoch | Loss | Epoch | Loss | Epoch | Loss | Epoch | Loss | Epoch | Loss |
|---|---|---|---|---|---|---|---|---|---|
| 202 | 117.269 | 262 | 108.835 | 322 | 108.323 | 382 | 108.246 | 442 | 108.231 |
| 204 | 116.416 | 264 | 108.792 | 324 | 108.314 | 384 | 108.244 | 444 | 108.232 |
| 206 | 115.616 | 266 | 108.755 | 326 | 108.307 | 386 | 108.244 | 446 | 108.235 |
| 208 | 114.884 | 268 | 108.729 | 328 | 108.300 | 388 | 108.244 | 448 | 108.234 |
| 210 | 114.228 | 270 | 108.696 | 330 | 108.296 | 390 | 108.242 | 450 | 108.239 |
| 212 | 113.651 | 272 | 108.666 | 332 | 108.290 | 392 | 108.240 | 452 | 108.234 |
| 214 | 113.171 | 274 | 108.639 | 334 | 108.291 | 394 | 108.242 | 454 | 108.236 |
| 216 | 112.740 | 276 | 108.610 | 336 | 108.279 | 396 | 108.240 | 456 | 108.234 |
| 218 | 112.354 | 278 | 108.593 | 338 | 108.280 | 398 | 108.238 | 458 | 108.237 |
| 220 | 112.010 | 280 | 108.565 | 340 | 108.281 | 400 | 108.237 | 460 | 108.236 |
| 222 | 111.688 | 282 | 108.554 | 342 | 108.277 | 402 | 108.238 | 462 | 108.234 |
| 224 | 111.403 | 284 | 108.532 | 344 | 108.276 | 404 | 108.244 | 464 | 108.236 |
| 226 | 111.130 | 286 | 108.523 | 346 | 108.270 | 406 | 108.240 | 466 | 108.234 |
| 228 | 110.880 | 288 | 108.496 | 348 | 108.266 | 408 | 108.233 | 468 | 108.234 |
| 230 | 110.646 | 290 | 108.483 | 350 | 108.262 | 410 | 108.231 | 470 | 108.232 |
| 232 | 110.426 | 292 | 108.464 | 352 | 108.260 | 412 | 108.232 | 472 | 108.240 |
| 234 | 110.238 | 294 | 108.450 | 354 | 108.259 | 414 | 108.236 | 474 | 108.240 |
| 236 | 110.072 | 296 | 108.433 | 356 | 108.258 | 416 | 108.233 | 476 | 108.239 |
| 238 | 109.917 | 298 | 108.423 | 358 | 108.252 | 418 | 108.234 | 478 | 108.242 |
| 240 | 109.771 | 300 | 108.404 | 360 | 108.252 | 420 | 108.234 | 480 | 108.235 |
| 242 | 109.636 | 302 | 108.398 | 362 | 108.254 | 422 | 108.236 | 482 | 108.231 |
| 244 | 109.512 | 304 | 108.381 | 364 | 108.254 | 424 | 108.233 | 484 | 108.234 |
| 246 | 109.401 | 306 | 108.373 | 366 | 108.251 | 426 | 108.236 | 486 | 108.233 |
| 248 | 109.301 | 308 | 108.362 | 368 | 108.250 | 428 | 108.233 | 488 | 108.234 |
| 250 | 109.208 | 310 | 108.349 | 370 | 108.246 | 430 | 108.232 | 490 | 108.228 |
| 252 | 109.125 | 312 | 108.342 | 372 | 108.253 | 432 | 108.235 | 492 | 108.234 |
| 254 | 109.056 | 314 | 108.336 | 374 | 108.244 | 434 | 108.234 | 494 | 108.232 |
| 256 | 108.990 | 316 | 108.334 | 376 | 108.245 | 436 | 108.231 | 496 | 108.232 |
| 258 | 108.932 | 318 | 108.324 | 378 | 108.248 | 438 | 108.232 | 498 | 108.235 |
| 260 | 108.880 | 320 | 108.323 | 380 | 108.245 | 440 | 108.230 | 500 | 108.232 |
From the above tables we can see that the proposed framework reaches stable convergence within a moderate number of epochs, and further training yields negligible improvements.
Thank you for the author's response; my questions have been resolved.
Reply 1
We thank reviewer jd1i for the valuable comments. Please find our response below.
1. Sensitive analysis of the hybrid prototype imputation strategy.
In Appendix G Figure 8 we analyze the performance of SPORT with different coefficient γ on all six datasets with 0.5 missing rate. From Figure 8 we can see that the performance of SPORT is relatively stable when γ is set from 0 to 0.4. For example, as shown in Figure 8 (c) on the Digit4k dataset, when γ varies between 0 and 0.4, the largest fluctuation in ACC is only by 2.40%. Meanwhile as shown in Figure 8 (d), on the 100Leaves dataset, the largest fluctuation in ACC is 4.2% as γ is set between 0 and 0.4. These results validate the robustness of SPORT's imputation strategy.
2. Discussions of the fixed coefficient for prototype imputation.
We agree that the performance of SPORT may vary due to different dataset characteristics by fixing γ. However, as shown in Figure 8 and discussed above, SPORT demonstrate a competitive and stable performance when γ varies from 0 to 0.4. This is because of the hybrid imputation strategy effectively balances prototype-level global structure and neighborhood-level local consistency. By integrating these two complementary sources of information, our model achieves more reliable reconstruction of missing views, which in turn leads to improved feature quality and more robust clustering performance across different datasets.
3. Discussion of relational weights under limited jointly-observed views.
We agree that the reliability of relational weights may decrease when only a limited number of views are jointly observed. Our work promotes the reliability under limited jointly-observed views in two aspects. First, instead of using a threshold to determine positive pairs based on similarity weights, we use soft weights smaller than 1 to capture structural relationships, thereby causing a smaller influence when structural relationships are unreliable. Second, the cross-view features corresponding to the same sample always have a weight of one, ensuring that the contrastive learning objective retains its basic cross-view alignment function even when inter-sample relational weights are less reliable. We will preserve further improving the reliability of structural information in future works
4. Explanation of the convergence analysis.
We thank reviewer jd1i for suggesting that a reporting the loss curve would be more sufficient for our convergence analysis. Previously we only report the performance because SPORT adopts a two-staged optimization framework where we use different losses during the pretraining and fine-tuning phase. Therefore, plotting the loss curve would seem a little disorganized due to the discontinuity between epochs 200 and 201. We present the convergence behavior of SPORT on Digit4k and 100Leaves with 0.5 missing rate in the following tables.