# The ICLR 6021 Rebuttal Pxr2
## Pxr2

### Weaknesses:
1. The quality Figure 2 needs to be improved.
2. Statistical results in tables will be more convincing. In addition, the optimal results in Table 1 should all be highlighted. (增加标准差)
### Questions：
1. What is the difference between Spatial RDA and Temporal RDA, and what roles do they play in this task?（增加flowchat，并额外强调）
2. Are the weights of each loss function in Formula 5 the same? Are the weights of each loss function considered?（增加权重符号，说明是被省略）

### Response to Reviewer Pxr2:
Thank you so much for your thoughtful comments and the time to provide constructive feedback to strengthen our work further! We have revised the paper to address your concerns as follows.

### Answer for Weakness 1:
**We have redrawn Figure 3 (i.e., Figure 2 in the original version) to improve the readability of the RDA module**. Specifically, we use different background colors to denote different sub-modules and thus make the figure details more transparent.

### Answer for Weakness 2:

In the tables of our original submission, we only report two types of statistical results, i.e., Accuracy and F1 Score. Following the reviewer's suggestion, we have added a new type of statistical results in Table 1. **We have added the standard deviation of all Accuracy and F1 Score**. Please see Table 1 in our revised PDF version. We have been conducting experiments to calculate the standard deviation of accuracy results for other tables.

To increase the readability of the results in Table 1, we have highlighted the best, second-best, and third-best results in red, green, and blue, respectively.

### Answer for Question 1:

**Structurally, Spatial RDA and Temporal RDA are the same; functionally, they infer and align relationship distributions of domains in the temporal and spatial dimensions, respectively**. Physiological signals are often interrelated in both time and space. Therefore, to fully infer and utilize this spatio-temporal relationship, we used a Spatial RDA and a Temporal RDA for the multi-modal physiological signals in different stages, respectively.

We have added a global flow chart in Figure 2 to illustrate the role of each module (e.g., Spatial RDA and Temporal RDA). We have also updated the description of the two RDAs in our paper as follows:

> RDA, as the core component of VBH-GNN, accepts node embeddings for domain alignment and updates the weights of node embeddings. The VBH-GNN contains structurally consistent Temporal RDA and Spatial RDA, which perform inference and alignment of relationship distributions in the temporal and spatial dimensions. The details of RDA will be explained in Section 3.2.

### Answer for Question 2:
**Yes, the weights of all the loss functions in Formula 5 are set to be the same; specifically, the weights are equal to one**.

We have conducted additional experiments to find the best weight combination. We found that using a weight of 1 for all terms achieves the best experimental results. Therefore, we omitted the weights symbolic in the previous submission version. Based on the reviewer's comment, we have added the weight symbols in Formula 5 for a better understanding. The revised Formula 5 is as follows:

> **Loss of VBH-GNN** contains two types of loss: the RDA Loss for aligning the source domain and target domain and the prediction loss of the classifier. The final loss function is formulated as
> $${\cal L}_{\text{VBH-GNN}} = \lambda_{1}{\cal L}_{\text{SRDA}}+\lambda_{2}{\cal L}_{\text{TRDA}}+ \lambda_{3}{\cal L}_{SBCE}+\lambda_{4}{\cal L}_{TBCE}$$ where ${\cal L}_{\text{SRDA}}$ and ${\cal L}_{\text{TRDA}}$ are loss of Spatial RDA and Temporal RDA (will be further explained in Section 3.2.3), ${\cal L}_{SBCE}$ and ${\cal L}_{TBCE}$ are Binary Cross-Entropy Loss for source and target domain classification. $\lambda_{1}$, $\lambda_{2}$, $\lambda_{3}$, and $\lambda_{4}$ are loss weights, which are all set to $1$ in the experiments.

We hope that our answers clarify your concerns. Thank you for your time and feedback! We are glad to have any further discussion.