- Rating
- 3(+1)
- Confidence
- 4
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper proposes FLO, a fast–slow reciprocal approach for multi-domain graph pre-training. This method mainly contain two components: PARR and FSKA. The former uses an attribute branch to predict relations between nodes without message passing and feeds these predictions back to the graph model. The latter uses fast and slow learning processes to reduce the cross-domain interference. Exiperiments are conducted in 12 datasets. The results show that FLO improves transfer performance.
Critical strengths
S1. The paper studies a meaningful problem: increasing the number of source domains does not always improve model performance.
S2. The method is clear and easy to understand.
S3. The experiments are extensive. The paper evaluates several different downstream tasks and provides comprehensive ablation studies.
Critical weaknesses
W1. One of the most important conclusion of the work is that FLO can benefits from more source domain, although the experiment only use six source doamin and use the fixed rank. I believe do more experiment about unfixed rank and more different domain can be more convincing.
W2. For a graph foundation model, one pre-training seed is too few to demonstrate robustness. I believe that five seeds for pre-training would be a sweet spot for supporting the competitive performance considering the cost.
W3. The hyperparameter settings are not consistent. In Appendix B, "Singular components kept for the target graph" is 16 for Cora, 8 for Citeseer, 512 for Computers and Propagation steps are 5 for Cora and 1 for other datasets .This is my biggest concern. Although experiment shows a strong competitive performance, additional track might affect the results significantly. I am curious about how much of the improvement actually comes from FLO's pre-training method?
Questions for authors
See W1~3
7 comments
R2 for W2: Five pre-training seeds
R2 for W2: Five pre-training seeds
Thank you for this valuable comment. We re-ran pre-training across five random seeds for FLO and the baselines in Table 3 across all three transfer tasks, and compare the results with those of the original submission. Table 3 reports the mean performance across the target datasets of each task, and the per-dataset five-seed results have been updated in the paper:
-
Node Classification: mean over Cora, CiteSeer, PubMed, CS, Photo, and Computers, corresponding to main Table 1 (per-dataset five-seed table in Appendix D.9, Table 19).
-
Link Prediction: mean over Physics, WikiCS, and DBLP, corresponding to main Table 2 (left) (per-dataset five-seed table in Appendix D.9, Table 20).
-
Graph Classification: mean over COIL-RAG, ENZYMES, and PROTEINS, corresponding to main Table 2 (right) (per-dataset five-seed table in Appendix D.9, Table 21).
Table 3: Multi-task transfer performance (%): node classification (NC, Acc.), link prediction (LP, Hits@50), and graph classification (GC, Acc.). Original is the single pre-training run in the original submission; 5 seeds is the mean ± std across five random pre-training seeds.
| Method | NC Original | NC 5 seeds | LP Original | LP 5 seeds | GC Original | GC 5 seeds |
|---|---|---|---|---|---|---|
| DGI | 82.60 | 82.79 ± 0.19 | — | 58.36 ± 0.34 | — | 63.19 ± 0.08 |
| GRACE | 78.83 | 78.68 ± 0.22 | — | 56.07 ± 0.83 | — | 63.69 ± 0.53 |
| BGRL | 83.69 | 83.88 ± 0.28 | — | 66.68 ± 0.58 | — | 63.77 ± 0.32 |
| MDGPT | 86.02 | 85.99 ± 0.10 | 57.69 | 58.17 ± 0.45 | 60.99 | 60.98 ± 0.21 |
| SAMGPT | 85.77 | 85.74 ± 0.10 | 59.18 | 58.31 ± 0.60 | 60.71 | 61.36 ± 0.59 |
| FUG | 84.96 | 85.60 ± 0.05 | 62.13 | 62.19 ± 0.48 | 63.34 | 63.21 ± 0.07 |
| TIG | 86.68 | 86.80 ± 0.03 | 58.75 | 59.60 ± 0.52 | 60.68 | 60.55 ± 0.22 |
| MDGFM | 85.87 | 85.92 ± 0.19 | 59.08 | 58.90 ± 0.42 | 62.71 | 63.27 ± 0.63 |
| BRIDGE | 85.80 | 85.78 ± 0.14 | 58.67 | 58.75 ± 0.27 | 61.55 | 61.50 ± 0.16 |
| FLO | 87.07 | 86.98 ± 0.06 | 71.13 | 71.84 ± 1.24 | 64.68 | 64.63 ± 0.20 |
"—": DGI, GRACE, and BGRL were added to the link prediction and graph classification comparison in the revision.
From Table 3, we draw two conclusions:
-
(i) FLO achieves the highest five-seed mean on all three tasks, outperforming the strongest baseline by 0.18 percentage points in node classification (TIG), 5.16 percentage points in link prediction (BGRL), and 0.86 percentage points in graph classification (BGRL).
-
(ii) The five-seed results agree with the original submission. FLO's mean changes by −0.09, +0.71, and −0.05 percentage points on the three tasks, and FLO ranks first on all three tasks in the original results, in the five-seed means, and on every individual seed. The cross-seed std of FLO is 0.06 percentage points in node classification.
R3 for W3: Shared few-shot configuration
R3 for W3: Shared few-shot configuration
Thank you for this valuable comment. To measure how much improvement comes purely from FLO's pre-training without per-dataset hyperparameter tuning, we re-evaluated FLO and the nine baselines in Table 4 under one unified configuration across all six target datasets:
- Single fixed SVD rank: FLO uses one shared SVD rank (21, chosen once on a separate validation subset) for every target graph, while baselines use their own native input adapters.
- Identical readout and propagation for all methods: Every method uses the same classifier and is evaluated both under Direct readout (0 propagation steps)—removing post-hoc graph propagation completely—and under 1 shared propagation step applied identically to all methods.
Table 4 reports the mean accuracy across the six target datasets (Appendix D.9, Table 22). The per-dataset results under the original setting of Figure 3 remain in Appendix D.4 (Tables 11–13).
Table 4: Six-target mean few-shot accuracy (%, mean ± std) under the shared configuration.
| Method | Direct (0-step) 1-shot | Direct (0-step) 3-shot | Direct (0-step) 5-shot | Shared (1-step) 1-shot | Shared (1-step) 3-shot | Shared (1-step) 5-shot |
|---|---|---|---|---|---|---|
| DGI | 47.97 ± 2.94 | 59.75 ± 2.11 | 64.98 ± 1.52 | 53.24 ± 3.48 | 65.38 ± 2.24 | 70.20 ± 1.54 |
| GRACE | 37.18 ± 2.46 | 48.14 ± 1.99 | 53.73 ± 1.66 | 43.18 ± 3.08 | 54.33 ± 2.46 | 59.49 ± 1.84 |
| BGRL | 45.47 ± 2.96 | 56.49 ± 2.30 | 61.84 ± 1.68 | 48.74 ± 3.15 | 59.48 ± 2.42 | 64.50 ± 1.79 |
| MDGPT | 52.46 ± 3.24 | 62.98 ± 2.25 | 67.58 ± 1.72 | 54.84 ± 3.47 | 65.05 ± 2.31 | 69.28 ± 1.74 |
| SAMGPT | 52.62 ± 3.35 | 63.13 ± 2.35 | 67.54 ± 1.80 | 55.29 ± 3.58 | 65.39 ± 2.42 | 69.35 ± 1.81 |
| FUG | 54.35 ± 3.39 | 64.80 ± 2.49 | 68.83 ± 1.98 | 57.23 ± 3.64 | 66.93 ± 2.60 | 70.45 ± 2.04 |
| TIG | 53.84 ± 3.78 | 59.96 ± 3.08 | 64.42 ± 2.73 | 54.62 ± 3.83 | 60.77 ± 3.15 | 65.21 ± 2.75 |
| MDGFM | 52.89 ± 3.15 | 64.28 ± 2.09 | 69.00 ± 1.50 | 55.38 ± 3.32 | 66.41 ± 2.12 | 70.74 ± 1.47 |
| BRIDGE | 50.93 ± 2.87 | 62.49 ± 2.01 | 67.65 ± 1.52 | 54.98 ± 3.21 | 66.10 ± 2.09 | 70.61 ± 1.51 |
| FLO | 57.17 ± 3.34 | 67.02 ± 2.25 | 70.43 ± 1.64 | 58.41 ± 3.47 | 67.74 ± 2.34 | 70.95 ± 1.69 |
From Table 4, we draw two conclusions:
-
(i) With direct readout (0 propagation steps) and a single SVD rank, FLO leads the strongest baseline by 2.82, 2.22, and 1.43 percentage points at 1, 3, and 5 shots, respectively.
-
(ii) With one shared propagation step applied equally to all methods, FLO still ranks first at every shot count (+1.18, +0.81, and +0.21 percentage points), showing that FLO's lead does not depend on per-dataset tuning.
R1 for W1: More source graphs, source order, and rank
R1 for W1: More source graphs, source order, and rank
Thank you for this valuable comment. To directly address your question, we conducted two sets of additional experiments: adding more source graphs (Table 1) and varying the order of source graphs (Table 2). All results and curves are included in Appendix D.9 of the revised manuscript (Tables 16–18 and Figures 10–11).
1. Adding more source graphs. Starting from the five source graphs in main Table 1, we sequentially add DBLP, Physics, and DBLP-APA, and evaluate all models with the same downstream evaluation as in main Table 1.
Table 1: Node classification accuracy (%, mean ± std) as new source graphs are sequentially added.
From Table 1, we draw two conclusions:
(i) FLO benefits from more source domains. As DBLP, Physics, and DBLP-APA are added, the six-target mean accuracy rises at every step: 87.01% → 87.02% → 87.18% → 87.35% (+0.34 percentage points overall, whereas the four multi-domain baselines gain at most 0.11 percentage points over the same expansion in Appendix Figure 10).
(ii) Five of the six individual targets improve, while the remaining one is already near saturation. CiteSeer (+1.12 percentage points) and Cora (+0.73 percentage points) gain the most. Computers already reaches 90.19% with five source graphs and remains stable within one standard deviation (90.13%).
2. Varying the source-domain order. We test domain order at both levels: (a) reversing the order in which the five source graphs are visited within each epoch (Table 2; Appendix D.9, Table 18), and (b) adding source graphs from one to five along six different inclusion orders (Appendix D.9, Figure 11 and Table 17).
Table 2: Node classification accuracy (%, mean ± std) under the original and reversed source-domain orders.
From Table 2 and Appendix Figure 11, we draw two conclusions:
(i) FLO is insensitive to source-domain order. Reversing the training order changes accuracy by at most 0.12 percentage points across CS, Photo, and Computers.
(ii) Averaged across six different source-inclusion orders, accuracy rises at every step as more source graphs are added (from 1 to 5 source graphs: 91.16% → 93.62% on CS, 89.37% → 93.62% on Photo, and 82.86% → 90.13% on Computers; see Appendix Figure 11 and Table 17).
3. Fixed versus unfixed rank. FLO ranks first both with per-target ranks (Appendix D.4, Tables 11–13) and with one shared rank of 21 for all six targets (Appendix D.9, Table 22; see R3 for W3).