- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This work proposes UGPA-AAD for spatial auditory attention decoding across different subjects, to discriminate left versus right attention purely from EEG signals. The framework consists of three core modules: MSFE, UGGE, and PCSL. The authors adopt leave-one-subject-out evaluation on DTU, MM-AAD, AVED and AV-GC datasets, using 1-second and 2-second time windows. Seven baseline methods are reimplemented for comparison.
Critical strengths
- Cross-subject AAD is practically valuable.
- Instead of directly copying numbers from existing papers, the authors re-run all baseline methods under a unified preprocessing and data partition pipeline.
- Weighting graph edges by uncertainty is an intuitive and reasonable design.
Critical weaknesses
- From what I can see, all seven baselines are trained only on source subjects. The paper also does not clarify whether these adaptation windows are exactly the same samples used for testing. The ablation results in the manuscript itself show how critical this issue is. When L_{tpcl} is removed (Table 7b): DTU: 56.2 / 56.9, lower than S2Former at 57.7 / 59.7. AV-GC: 57.8 at 1 s, nearly on par with PCRNet (57.9); 57.6 at 2 s, roughly +1.0 above the best baseline MHANet (56.6). So most performance advantage reported over baselines likely comes from access to target-subject data, rather than the UGGE module or prototype design.
- Accuracy values stay close to random guess level, and statistical significance tests are missing.I ask the authors to supply: paired subject-wise significance tests (e.g. Wilcoxon test) against the strongest baseline, per-subject accuracy records, and performance variance across different random seeds.
- For AV-GC, gaze direction is a well-known confounding factor. The paper notes that its strongest learned graph connections are fronto-central channels (Fig.7). The largest performance gains appear in audiovisual settings: AV-GC (+2.7 / +3.8), AVED-AV (+3.3 / +3.1). Taken together, these observations suggest the improvement may come from eye movement signals rather than auditory attention itself.
- Hyperparameter selection and checkpoint picking appear to rely on test-set performance. Please clarify this.
- Several inconsistencies exist in method description. For examples, "SA" is ablated in Table3 but never defined; Equation (14) computes cross entropy over f_m, while Section3.1 states the classifier operates on z. Please clarify which feature feeds the classifier, and how the classification head on z is optimized; The paper says \sigma is refined into \hat{\sigma}, but Equation (8) only updates \mu.
Questions for authors
- Are the unlabeled target windows used for adaptation exactly the same samples used for testing? What are the performance of all methods without target data access?
- Can you share per-subject accuracy, paired significance test against the best baseline, and performance variance across random seeds?
- What are AV-GC results split by condition, especially gaze-incongruent trials? Was any ocular artifact removal applied? Why were 6 out of 16 participants excluded?
- Is there empirical evidence showing that \sigma indeed captures subject-wise or channel-wise signal unreliability?
4 comments
Response - Part 1
We thank the reviewer for the careful assessment and constructive requests. We agree that data access, subject-level reliability, potential ocular confounds, and implementation consistency must be made explicit. We address each concern below and will incorporate the corresponding clarifications and verified analyses into the revision.
1. Data Access, Adaptation Fairness, and the Role of TPCL (W1, Q1)
We respectfully clarify that UGPA-AAD and all seven baselines draw their data from the same pool of K non-test subjects in each outer leave-one-subject-out (LOSO) fold, while the held-out test subject is reserved exclusively for final evaluation. As defined in Sec. 3.1, each fold contains K+1 subjects, comprising K−1 source subjects, one adaptation subject, and one held-out test subject (K−1:1:1). The adaptation subject is randomly selected from the K non-test subjects, with 60% of its trials used for unlabeled TPCL adaptation and the remaining 40% for labeled validation. The seven original baselines are trained using labeled data from K non-test subjects, whereas UGPA-AAD uses K−1 labeled source subjects and the designated adaptation subject. All methods follow the same LOSO evaluation protocol across all datasets. We have clarified these important details and the ambiguous use of "target subject" in the revised manuscript, with changes highlighted in blue.
To assess adaptation effectiveness under matched data access, we further compare representative baselines augmented with DANN, Deep CORAL, pseudo-label self-training (PL), and prototype alignment (PA), using identical source, adaptation, and validation partitions. The results are presented in Table R1. UGPA-AAD consistently outperforms all adaptation-enhanced baselines, indicating that its gains are not solely attributable to access to unlabeled adaptation data. Furthermore, the w/o TPCL results and Table 3 ablations support the complementary contributions of TPCL, UGGE, and PCSL.
Table R1. Adaptation comparison under matched data access (Acc.% ± SD).
| Method | DTU 1s | DTU 2s | AV-GC 1s | AV-GC 2s |
|---|---|---|---|---|
| S²M-Former | 57.7 ± 4.21 | 59.7 ± 5.25 | 54.0 ± 3.78 | 53.2 ± 3.89 |
| S²M-Former+DANN | 52.8 ± 8.43 | 54.6 ± 7.92 | 51.7 ± 6.81 | 50.9 ± 8.17 |
| S²M-Former+CORAL | 54.3 ± 9.26 | 55.1 ± 8.74 | 50.8 ± 7.63 | 52.4 ± 7.19 |
| S²M-Former+PL | 53.6 ± 7.58 | 55.9 ± 6.93 | 54.6 ± 5.82 | 54.1 ± 6.47 |
| S²M-Former+PA | 55.2 ± 8.17 | 53.8 ± 9.12 | 51.4 ± 8.36 | 52.3 ± 7.84 |
| ListenNet | 56.8 ± 7.32 | 57.2 ± 5.83 | 53.2 ± 4.90 | 55.0 ± 7.06 |
| ListenNet+DANN | 51.2 ± 9.06 | 51.9 ± 8.97 | 50.4 ± 0.12 | 50.0 ± 8.09 |
| Listennet+CORAL | 53.1 ± 10.17 | 53.2 ± 9.79 | 49.9 ± 8.22 | 50.1 ± 7.81 |
| ListenNet+PL | 50.5 ± 9.60 | 51.1 ± 9.45 | 52.4 ± 5.47 | 53.0 ± 5.06 |
| ListenNet+PA | 52.1 ± 9.16 | 53.2 ± 9.81 | 49.8 ± 8.93 | 50.2 ± 8.77 |
| UGPA-AAD w/o TPCL | 56.2 ±4.15 | 56.9 ±3.81 | 57.8 ±8.16 | 57.6 ±9.18 |
| UGPA-AAD | 58.8 ± 3.73 | 59.2 ± 3.44 | 60.6 ± 7.58 | 60.4 ± 8.68 |
2. Eye-Gaze Confounds and AV-GC Participant Selection (W3, Q3)
We thank the reviewer for raising the important concern regarding potential gaze-related confounds in AV-GC. Our study focuses on cross-subject auditory attention decoding under a unified leave-one-subject-out (LOSO) protocol. Specifically, all four experimental conditions of AV-GC are included in the evaluation of each held-out test subject, and the reported accuracy reflects the overall decoding performance across these conditions. No condition-specific models are trained, and gaze-direction labels are not used as inputs or supervision during training or adaptation.
The consistent improvements on auditory-only datasets (DTU and MM-AAD) demonstrate the effectiveness of our method beyond audiovisual settings. Moreover, Fig. 7 reveals prominent connections across fronto-central, temporal, and parietal regions, suggesting that the learned representations are not restricted to regions potentially associated with ocular activity.
For AV-GC, we followed the official preprocessing pipeline and S²Former's experimental settings without additional ocular-artifact removal. As explained in Appendix A.1 (Dataset Details), we retained 10 participants with complete recordings across all four conditions. Specifically, participants 2, 5, and 6 were excluded due to consent restrictions, while participants 1, 3, and 14 had incomplete recordings, consistent with the official dataset documentation.
Response - Part2
3. Subject-Wise Significance and Random-Seed Robustness (W2, Q2)
We conducted a subject-wise comparison between UGPA-AAD and the strongest baseline, ListenNet, on AVED-AV using a two-sided exact Wilcoxon signed-rank test. UGPA-AAD achieves mean improvements of 3.22% and 3.18% for 1 s and 2 s windows, respectively, with statistically significant differences (p = 0.005859 and p = 0.013672). Furthermore, UGPA-AAD outperforms ListenNet on 9 of 10 subjects in both settings, with significance maintained after Holm correction (adjusted p = 0.011719 and 0.013672).
Table R2. Subject-wise classification accuracy and paired tests on AVED-AV (%).
| Subject | UGPA-AAD 1 s | ListenNet 1 s | UGPA-AAD 2 s | ListenNet 2 s |
|---|---|---|---|---|
| S01 | 55.05 | 53.61 | 55.42 | 53.29 |
| S02 | 55.01 | 53.72 | 55.22 | 53.70 |
| S03 | 59.73 | 56.53 | 60.31 | 53.83 |
| S04 | 58.03 | 52.74 | 59.35 | 53.98 |
| S05 | 52.40 | 53.40 | 52.61 | 52.24 |
| S06 | 60.68 | 52.38 | 64.20 | 52.65 |
| S07 | 59.52 | 52.38 | 59.64 | 54.97 |
| S08 | 57.16 | 54.56 | 55.71 | 55.32 |
| S09 | 58.82 | 55.53 | 57.00 | 56.47 |
| S10 | 53.23 | 52.59 | 52.94 | 54.14 |
| Mean | 56.96 | 53.74 | 57.24 | 54.06 |
| Wilcoxon W | 2.0 | 4.0 | ||
| Two-sided p | 0.005859 | 0.013672 |
To assess robustness to random initialization, we conducted experiments with three random seeds (42, 1024, and 3407) across all datasets and both window lengths. As shown in Tables R3 and R4, the standard deviation of mean accuracy across seeds ranges from 0.12% to 0.46%, demonstrating consistent performance across random initializations.
Table R3. Performance across random seeds on DTU and MM-AAD (Acc.% ± SD).
| Seed | DTU 1 s | DTU 2 s | MM-AAD AO 1 s | MM-AAD AO 2 s | MM-AAD AV 1 s | MM-AAD AV 2 s |
|---|---|---|---|---|---|---|
| 42 | 58.8 ± 3.73 | 59.2 ± 3.44 | 58.1 ± 5.42 | 58.5 ± 5.18 | 57.0 ± 4.35 | 57.2 ± 4.26 |
| 1024 | 58.4 ± 4.36 | 59.0 ± 3.85 | 58.2 ± 5.13 | 58.7 ± 5.29 | 57.4 ± 4.97 | 57.5 ± 4.33 |
| 3407 | 59.1 ± 3.82 | 59.6 ± 3.17 | 57.9 ± 4.91 | 58.5 ± 5.36 | 57.6 ± 4.23 | 57.6 ± 3.90 |
| Mean | 58.77 | 59.27 | 58.07 | 58.57 | 57.33 | 57.43 |
Table R4. Performance across random seeds on AVED and AV-GC (Acc.% ± SD).
| Seed | AVED AO 1 s | AVED AO 2 s | AVED AV 1 s | AVED AV 2 s | AV-GC 1 s | AV-GC 2 s |
|---|---|---|---|---|---|---|
| 42 | 54.1 ± 1.25 | 54.0 ± 1.44 | 57.0 ± 2.74 | 57.2 ± 3.43 | 60.6 ± 7.58 | 60.4 ± 8.68 |
| 1024 | 54.4 ± 2.31 | 54.2 ± 1.36 | 56.8 ± 2.96 | 57.3 ± 3.33 | 60.0 ± 7.19 | 60.0 ± 8.31 |
| 3407 | 53.7 ± 1.38 | 53.7 ± 1.22 | 57.0 ± 2.70 | 56.9 ± 3.11 | 59.7 ± 6.14 | 59.9 ± 7.57 |
| Mean | 54.07 | 53.97 | 56.93 | 57.13 | 60.10 | 60.10 |
4. Hyperparameters and Checkpoint Selection (W4)
No test-subject data or test accuracy was used for hyperparameter tuning or checkpoint selection. Hyperparameters were determined using the training set, with key settings detailed in Sec. 4.1. In each LOSO fold, the optimal checkpoint was selected based on the lowest supervised loss on the 40% labeled validation trials of the non-test adaptation subject. The remaining 60% unlabeled trials were used for TPCL adaptation, and the held-out test subject was evaluated only after checkpoint selection.
5. Empirical Support for σ as a Reliability Estimate (Q4)
Within our Gaussian representation framework, σ models feature-level uncertainty that may reflect inter-subject variability. As illustrated in Fig. 1(b), σ exhibits more pronounced variations across subjects than μ, highlighting its sensitivity to subject-dependent feature differences. The w/o RGL ablation in Appendix B.1 (Table 5) demonstrates the contribution of reliable graph learning. To further isolate the role of σ, we introduce a μ-only graph variant under the same architecture and training settings. As shown in Table R5, UGGE consistently outperforms this variant across all four settings, with accuracy gains of 0.90% to 1.7%.
Table R5. Graph construction comparison (Acc.% ± SD).
| Graph | DTU 1s | DTU 2s | AV-GC 1s | AV-GC 2s |
|---|---|---|---|---|
| Random Graph | 58.0 ± 3.86 | 58.4 ± 3.59 | 59.5 ± 7.78 | 59.6 ± 8.86 |
| -only | 57.9 ± 3.99 | 58.1 ± 4.12 | 58.9 ± 7.34 | 58.8 ± 9.25 |
| UGGE | 58.8 ± 3.73 | 59.2 ± 3.44 | 60.6 ± 7.58 | 60.4 ± 8.68 |
Response - Part3
6. Notation, Classification Features, and Uncertainty Updates (W5)
We thank the reviewer for identifying these presentation inconsistencies. We clarify the following points and have revised the corresponding descriptions for consistency:
-
SA in Table 3: SA denotes the subject-invariance learning module, which employs an adversarial discriminator to suppress subject-specific information and encourage subject-invariant representations. Its objective is defined in Eq. (12), with details provided in L270–284 and the module architecture illustrated in the lower-right part of Fig. 2.
-
Classification features (Eq. 14): The classification loss in Eq. (14) is applied to , rather than . The final linear classifier is trained jointly with the feature extractor through . We have corrected Eq. (14) and the corresponding formulations to ensure consistency throughout the manuscript. The labeled adaptation trials are used only for checkpoint selection, not classifier training (see our response to W1 & Q1 for details).
-
Uncertainty updates (Eq. 8): We apologize for the ambiguity in our description. As shown in Eq. (8), only is directly refined by the GCN, while characterizes uncertainty and determines node reliability for graph construction, thereby influencing the adjacency matrix and subsequent updates of . Both Gaussian parameters remain trainable through end-to-end optimization with KL regularization. We have revised the description of to distinguish parameter learning from graph refinement.
Concluding Remarks
We sincerely thank the reviewer again for the encouraging assessment and constructive feedback. We hope that the additional graph construction analysis, clarified experimental protocol, and subject-wise statistical comparisons address your main concerns and provide stronger evidence for the effectiveness and reliability of UGPA-AAD.
The authors have not responded to my comments. Therefore, I keep my original rating.