- Rating
- 2
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper investigates the representation-to-competence gap in EEG foundation models, where physiologically meaningful information encoded within pretrained representations may not be fully utilized by downstream classifiers. Motivated by the observation that certain descriptors are more accessible at internal layers than at the final classifier input, they investigate low-rank residual adaptation at different backbone depths while keeping the pretrained weights and classifier frozen. Experiments across 4 backbones and 4 tasks demonstrate that internal residual adaptation generally outperforms final-output adaptation under matched parameter budgets.
Critical strengths
1.Connecting physiological representation analysis with adaptation placement provides an interesting perspective
2.The paper compares internal and final-output residual adaptation under matched trainable parameter budgets
Critical weaknesses
1.Although SCOPE and NeuroTTT are cited, the proposed internal residual adaptation shares similarities with SCOPE's internal adapter strategy, while NeuroTTT also explores physiologically motivated adaptation objectives and LoRA-based adaptation. The authors should more clearly distinguish their contributions from these existing approaches.
2.The paper motivates internal adaptation using the observation that physiological descriptors are more accessible at certain internal layers. However, the proposed residual adapters are trained solely with downstream task labels, and their placement is selected through task validation rather than physiological readout scores.
3.For CBraMod and CSBrain, the authors show that final-output residual adaptation remains within the affine function class of the reference classifier, whereas internal adaptation benefits from additional nonlinear backbone transformations. Comparisons with nonlinear classification heads and established parameter-efficient fine-tuning strategies would be helpful.
4.Several reported improvements are relatively small, so statistical significance tests or confidence intervals would better show the significance of the performance.
Questions for authors
1.Could the authors further clarify the methodological differences with existing methods such as NeuroTTT and SCOPE?
2.Could the authors quantitatively examine whether physiological readout scores at different layers correlate with the downstream improvements obtained by adapting those layers?
6 comments
Response to Reviewer jNpL (2/5), W2
W2. Training with task labels alone, and placement by task validation (Tables 18 and 19). The residual is indeed trained with the task loss alone. For placement, however, the two studies of the paper differ. In the location study of Table 1, we assigned the sites before training with a rule that reads D and the readout profile (Appendix A.7). In the depth comparison, the site is selected by task validation. The physiological analyses contribute as measurements at three points. Before adaptation, D names the family that the classifier at POST lacks, and E shows where this family can be read out. For LaBraM and CBraMod, the readout is higher at PRE or MID than at POST in 26 of 32 combinations. After adaptation, D of that family falls in the three settings where the residual raises Macro-F1 by more than 2 points (from 4.38 to 1.86 for CBraMod on ADFTD, from 2.57 to 1.15 for CBraMod on CAUEEG, and from 3.57 to 2.90 for LaBraM on ADFTD), and its readout at POST rises (Tables 18 and 19).
Table 18. Incremental task value D at POST before and after internal adaptation, for the family with the largest development value in Table 9. Each entry gives Macro-F1 / balanced accuracy differences (%). Gain is the assigned-site residual minus the reference. Adapted D is the mean ± SD across three seeds. Settings are ordered by Macro-F1 gain.
| Model | Task | Family | Gain | Frozen D | Adapted D |
|---|---|---|---|---|---|
| CBraMod | ADFTD | A | +3.45 / +2.90 | +4.38 / +3.88 | +1.86 ± 0.30 / +1.87 ± 0.29 |
| CBraMod | CAUEEG | R | +2.55 / +2.37 | +2.57 / +2.46 | +1.15 ± 0.19 / +0.96 ± 0.22 |
| LaBraM | ADFTD | A | +2.27 / +1.58 | +3.57 / +2.74 | +2.90 ± 0.25 / +2.38 ± 0.16 |
| LaBraM | TUSZ | R | +0.91 / +1.21 | +2.94 / +2.64 | +2.75 ± 0.18 / +2.46 ± 0.15 |
| CBraMod | TUSZ | R | +0.78 / +0.57 | +5.66 / +4.98 | +4.65 ± 0.38 / +4.33 ± 0.43 |
| BrainOmni | CAUEEG | R | +0.73 / +0.70 | -0.08 / -0.10 | +0.21 ± 0.83 / +0.23 ± 0.82 |
| CBraMod | TD-BRAIN | R | +0.61 / +0.82 | +5.28 / +5.50 | +4.81 ± 0.22 / +4.97 ± 0.25 |
| LaBraM | TD-BRAIN | A | +0.48 / +0.73 | +4.22 / +4.44 | +4.22 ± 0.21 / +4.43 ± 0.19 |
| BrainOmni | TUSZ | R | +0.37 / +0.38 | +1.88 / +1.77 | +1.52 ± 0.04 / +1.43 ± 0.03 |
| BrainOmni | ADFTD | A | +0.32 / +0.31 | -0.49 / -0.47 | -0.42 ± 0.06 / -0.38 ± 0.06 |
| BrainOmni | TD-BRAIN | R | +0.27 / +0.18 | +2.37 / +2.05 | +2.45 ± 0.10 / +2.11 ± 0.11 |
| LaBraM | CAUEEG | R | -0.13 / -0.21 | +0.57 / +0.44 | +0.91 ± 0.13 / +0.71 ± 0.16 |
Table 19. Task scores, incremental task value D, and POST readout for the reference, the assigned-site residual, and LoRA in all attention blocks. Task scores and D are given as Macro-F1 (F1) and balanced accuracy (BAcc). Readout is the ridge readout at POST of the descriptors after subtraction of their class means on the training subjects. Params is the number of trainable parameters. Adapted entries are the mean ± SD across three seeds.
| Adaptation | Params | Score F1 (%) | Score BAcc (%) | D F1 (%) | D BAcc (%) | Readout |
|---|---|---|---|---|---|---|
| LaBraM ADFTD, family A | ||||||
| Frozen | 0 | 45.43 | 46.44 | +3.57 | +2.74 | 0.1011 |
| Single site residual | 3,200 | 47.70 ± 0.31 | 48.01 ± 0.24 | +2.90 ± 0.25 | +2.38 ± 0.16 | 0.1104 ± 0.0023 |
| All block LoRA | 154,203 | 51.13 ± 3.01 | 51.24 ± 3.18 | -0.32 ± 0.70 | -0.31 ± 0.67 | 0.1184 ± 0.0069 |
| CBraMod ADFTD, family A | ||||||
| Frozen | 0 | 44.05 | 44.77 | +4.38 | +3.88 | 0.0907 |
| Single site residual | 3,200 | 47.51 ± 0.56 | 47.67 ± 0.51 | +1.86 ± 0.30 | +1.87 ± 0.29 | 0.0968 ± 0.0001 |
| All block LoRA | 154,203 | 53.17 ± 0.44 | 53.12 ± 0.51 | +0.45 ± 0.16 | +0.48 ± 0.17 | 0.1079 ± 0.0010 |
Response to Reviewer jNpL (3/5), W3 nonlinear heads
W3. Comparison with nonlinear classification heads (Tables 23 and 24). We trained an MLP classifier on the frozen POST features with the parameter count of the residual. The residual outperforms it in 7 of 8 settings for LaBraM and CBraMod (Table 24). We also added a FINAL residual with a GELU between its two matrices, which lies outside the affine class of Proposition A.8 and covers all 16 settings, including those of CSBrain (Table 23). It differs from the linear FINAL residual by at most 0.21 points in 15 of them, and the internal residual outperforms both in 12 of 14.
Table 24. MLP classifier on frozen POST features at the parameter count of the internal residual. Each entry gives pooled-window Macro-F1 / balanced accuracy (%), averaged across three seeds.
| Model | Task | Reference | MLP classifier | Assigned site |
|---|---|---|---|---|
| LaBraM | TD-BRAIN | 63.54 / 62.20 | 60.75 / 59.11 | 64.03 / 62.93 |
| LaBraM | ADFTD | 45.43 / 46.44 | 45.57 / 46.37 | 47.70 / 48.01 |
| LaBraM | CAUEEG | 51.50 / 52.67 | 51.05 / 51.63 | 51.37 / 52.45 |
| LaBraM | TUSZ | 44.58 / 44.57 | 42.93 / 43.12 | 45.49 / 45.77 |
| CBraMod | TD-BRAIN | 60.00 / 58.65 | 57.84 / 56.29 | 60.61 / 59.47 |
| CBraMod | ADFTD | 44.05 / 44.77 | 45.23 / 46.23 | 47.51 / 47.67 |
| CBraMod | CAUEEG | 49.69 / 51.13 | 48.88 / 50.41 | 52.24 / 53.50 |
| CBraMod | TUSZ | 42.43 / 42.92 | 42.51 / 43.65 | 43.22 / 43.48 |
Table 23. FINAL controls and classification from frozen internal features. Each entry gives pooled-window Macro-F1 / balanced accuracy (%), averaged across three seeds. Reference, assigned site, and FINAL linear follow Table 1, and † marks an assigned site at FINAL. FINAL GELU places a GELU between the two matrices of the FINAL residual. The internal classifier is a logistic classifier on frozen features of one site among PRE, MID, and FINAL, selected by task validation.
| Model | Task | Reference | Assigned site | FINAL linear | FINAL GELU | Internal classifier |
|---|---|---|---|---|---|---|
| LaBraM | TD-BRAIN | 63.54 / 62.20 | 64.03 / 62.93 | 63.42 / 62.16 | 63.51 / 62.28 | 64.93 / 63.67 |
| LaBraM | ADFTD | 45.43 / 46.44 | 47.70 / 48.01 | 46.09 / 46.89 | 45.98 / 46.82 | 44.33 / 45.61 |
| LaBraM | CAUEEG | 51.50 / 52.67 | 51.37 / 52.45 | 51.33 / 52.35 | 51.47 / 52.45 | 52.33 / 53.14 |
| LaBraM | TUSZ | 44.58 / 44.57 | 45.49 / 45.77 | 45.06 / 45.15 | 44.86 / 44.95 | 44.52 / 44.80 |
| CBraMod | TD-BRAIN | 60.00 / 58.65 | 60.61 / 59.47 | 60.01 / 58.70 | 59.95 / 58.62 | 64.04 / 62.68 |
| CBraMod | ADFTD | 44.05 / 44.77 | 47.51 / 47.67 | 44.02 / 44.74 | 44.02 / 44.75 | 47.21 / 48.22 |
| CBraMod | CAUEEG | 49.69 / 51.13 | 52.24 / 53.50 | 50.11 / 51.40 | 49.98 / 51.34 | 51.89 / 53.26 |
| CBraMod | TUSZ | 42.43 / 42.92 | 43.22 / 43.48 | 42.69 / 43.14 | 42.70 / 43.18 | 38.10 / 39.35 |
| CSBrain | TD-BRAIN† | 52.82 / 52.75 | 52.85 / 52.77 | 52.85 / 52.77 | 52.86 / 52.76 | 64.03 / 62.64 |
| CSBrain | ADFTD | 48.79 / 48.71 | 48.93 / 48.85 | 48.73 / 48.65 | 48.79 / 48.71 | 42.57 / 44.84 |
| CSBrain | CAUEEG† | 43.97 / 44.11 | 44.01 / 44.17 | 44.01 / 44.17 | 43.93 / 44.07 | 48.90 / 50.58 |
| CSBrain | TUSZ | 37.42 / 37.50 | 36.41 / 37.19 | 34.94 / 36.73 | 35.72 / 37.56 | 38.05 / 39.43 |
| BrainOmni | TD-BRAIN | 55.80 / 55.70 | 56.06 / 55.88 | 56.06 / 55.80 | 55.97 / 55.76 | 53.62 / 52.56 |
| BrainOmni | ADFTD | 52.68 / 52.56 | 53.00 / 52.87 | 52.73 / 52.64 | 52.73 / 52.63 | 49.86 / 51.57 |
| BrainOmni | CAUEEG | 50.15 / 50.88 | 50.88 / 51.58 | 50.35 / 50.91 | 50.17 / 50.89 | 53.08 / 54.29 |
| BrainOmni | TUSZ | 38.90 / 38.93 | 39.27 / 39.31 | 39.03 / 39.05 | 38.91 / 38.94 | 41.94 / 42.54 |
Response to Reviewer jNpL (1/5), Q1, W1 and Q2
We thank the reviewer. We first answer the two questions and then address the weaknesses in order. The answers refer to the revised manuscript (Section 4.6 and Appendix G). This response has five parts. Each table is reproduced below the answer that uses it, under its number in the revised manuscript.
Q1, W1. Methodological differences from SCOPE and NeuroTTT. The paper measures how much task information the frozen output leaves inside pretrained EEG backbones, and we use the residual as the instrument of this measurement. SCOPE addresses settings with few labels. It freezes the backbone and conditions modules in its deeper layers on prototypes. In this paper, we train a single linear residual from the task labels and study the effect of its site, from PRE to FINAL at a fixed budget of 3,200 parameters. The site has a large effect. For CBraMod on ADFTD, the gains are 3.1 to 4.4 points at blocks 1 to 6 and vanish at blocks 11 and 12 (Table 28, reproduced under Q2).
NeuroTTT trains the whole backbone with the supervised loss and self-supervised objectives and then adapts at test time, with LoRA across blocks as a baseline. In this paper, the backbone and the classifier stay frozen, the task loss is the only training signal, and the physiological descriptors are used as measurements. Used as auxiliary objectives, the descriptors stay below adaptation with the task loss alone in all 12 comparisons (Appendix F). We now compare with LoRA and bottleneck adapters in all blocks in Table 21, reproduced under W3.
Q2. Do readout scores at different layers correlate with the gains from adapting those layers? (Tables 27 and 28) We trained residuals at 13 sites of CBraMod on ADFTD, namely PRE and each of the 12 blocks (Table 28). Across the 13 sites, the Spearman correlation between the frozen readout and the gain is 0.57 for the smoothed spectral family and 0.82 for the morphology family, the two families with a positive D on this task. It is 0.49 and 0.02 for the two families with a negative D. We also trained residuals at PRE, MID and FINAL in 15 settings (Table 27). The best site is an internal one in all 15, and in 12 of them the readout of the family with the largest D on the development folds is also higher at an internal site than at FINAL. Both the readout and the gain are higher at internal sites.
Table 28. Residuals at 13 sites of CBraMod on ADFTD. Scores are pooled-window Macro-F1 (F1) and balanced accuracy (BAcc), reported as mean ± SD across three seeds (%). Gains are relative to the reference. Task validation selection chooses among the 13 sites within each fold and seed.
Table 27. Residuals at PRE, MID, and FINAL and task validation selection among the three sites. Each entry gives pooled-window Macro-F1 / balanced accuracy (%), averaged across three seeds. Selection is made within each fold and seed.