Diffusion Probabilistic Models for Structured Node Classification

This paper studies structured node classification on graphs, where the predictions should consider dependencies between the node labels. In particular, we focus on solving the problem for partially labeled graphs where it is essential to incorporate the information in the known label for predicting the unknown labels. To address this issue, we propose a novel framework leveraging the diffusion probabilistic model for structured node classification (DPM-SNC). At the heart of our framework is the extraordinary capability of DPM-SNC to (a) learn a joint distribution over the labels with an expressive reverse diffusion process and (b) make predictions conditioned on the known labels utilizing manifold-constrained sampling. Since the DPMs lack training algorithms for partially labeled data, we design a novel training algorithm to apply DPMs, maximizing a new variational lower bound. We also theoretically analyze how DPMs benefit node classification by enhancing the expressive power of GNNs based on proposing AGG-WL, which is strictly more powerful than the classic 1-WL test. We extensively verify the superiority of our DPM-SNC in diverse scenarios, which include not only the transductive setting on partially labeled graphs but also the inductive setting and unlabeled graphs.

Paper

Similar papers

Peer review

Reviewer w4tc5/10 · confidence 3/52023-07-05

Summary

This paper focuses on structured node classification on graphs by integrating diffusion probabilistic models to generate label distributions for unknown nodes. The authors design a novel training algorithm to apply DPMs, and a new variational lower bound. Theoretical analysis proves the proposed model is strictly more powerful than the classic 1-WL test. Experimental results also show its superiority over baseline methods on both inductive and transductive settings.

Strengths

- The paper is well written and structured. - Application of diffusion problistic models in node classification is novel. - The theoredical analysis of 1-WL test is convincing.

Weaknesses

- The use of diffusion problistic model is not well motivated? The reason for using diffusion process for partially labeled graphs is not explained. - It lacks a analysis of computational complexity of DPM-SNC, since it requires multiple steps of sampling.

Questions

- Please explain the reason of introducing diffusion problistic models. How can diffusion process improves GNNs on partially labeled graphs? - What is the computational complexity of DPM-SNC?

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

Please refer to the weaknesses.

Reviewer 7A8R7/10 · confidence 4/52023-07-07

Summary

The paper addresses the issue of structured node classification on graphs that consider dependencies between node labels. While GNNs have shown great success in machine learning on graphs, they predict node labels independently. The authors propose a novel framework, DPM-SNC (Diffusion Probabilistic Models for Structured Node Classification), in which they use DPMs to learn a joint label distribution over both known and unknown labels while making predictions based on known labels using manifold-constrained sampling. Additionally, they present a novel training algorithm for DPMs. Theoretical analysis shows that it enhances the expressive power of GNNs. Experiments across multiple settings demonstrate the effectiveness of the proposed approach.

Strengths

**Originality** This paper ... - combined GNNs and DPMs for structured node classification. - proposed using manifold-constraint sampling for the transductive setting. - gave a novel expressivity metric, AGG-WL and proved that DPM-SNC is more powerful than its backbone 1-WL-GNN. **Quality** In general, the math is solid and the claims are intuitive and/or empirically validated. The experiments are extensive. **Clarity** The paper is well-written and easy to follow. **Significance** The paper makes interesting and inspiring theoretical connections. Realistically, I am not so sure that this technique would be adopted in a production setting considering its cost and complexity.

Weaknesses

No major flaws. See questions.

Questions

1. What are the training and inference costs of the proposed method? How do they compare to the baselines? 2. Regarding model training, do you use the same objective as DDPM to optimize the GNN parameters? Based on my understanding of your paper and code (correct me if I am wrong), the $q(y_U|y_L)$ term in the objective is not used to update $\theta$. Instead, it is optimized by updating the buffer $\mathcal{B}$. Maybe the "Training algorithm" paragraph could be revised so that it better reflects the proposed method? 3. Among all kinds of generative models, why do you choose DPMs?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

I do not see any significant, unreported negative societal impact.

Reviewer JeF86/10 · confidence 4/52023-07-07

Summary

This paper proposes a diffusion probability model learning method for structured node classification, named DPM-SNC. The contributions of this paper are as follows. First, this paper leverages the high capacity of DPMs in learning joint label dependency and predicting posterior distribution conditioned on partially known data by devising DPM-SNC. Second, this paper introduces a new variational lower bound to train a DPM on partially labelled graphs. Moreover, this paper proposes AGG-WL test to theoretically analyse the expressive power of DPM-SNC using 1-WL-GNNs, which is more powerful than 1-WL-GNNs. Finally, detailed experiments on both transductive and inductive settings demonstrate DPM-SNC’s impressive performance.

Strengths

There are strengths of proposed method: 1. Applying DPM to learning partially labeled node label, to the best of our knowledge, is novel. 2. The experiments are detailed analyzed with both synthetic and real world data, which points out the strength of the method in different scenarios. The algorithmic reasoning shows the influence of different algorithmic elements in DPM, with great performance compared with baseline.

Weaknesses

However, there are several drawbacks: 1. The idea to apply DPM to learn the joint distribution over the labels is novel. However, this paper has little novelty to DPM itself, which utilizes manifold-constrained sampling from previous works. 2. The analysis for expressive power follows the color assignments description in 1-WL test. The details of AGG-WL, the perturbations, and the aggregation need more explanation.

Questions

1. Add more explanation to AGG-WL.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

2 fair

Limitations

Authors have adequately addressed the limitations in this paper.

Reviewer kJkj5/10 · confidence 3/52023-07-07

Summary

This paper investigates the use of diffusion probabilistic models (DPM) to model node-wise label dependencies for structured node classification. The proposed DPM-SNC generalizes diffusion probabilistic models to work on partially labeled graphs in the transduction settings. The proposed model is proved to be more expressive than the classic 1-WL test. Experimental results demonstrate the effectiveness of the proposed model.

Strengths

1. This paper leverages the merits of diffusion probabilistic models (DPM) for structured node classification. A new training algorithm is proposed, which generalises DPM to work on partially labeled graphs. 2. The theoretical analysis proves the improved expressive power of the proposed model. 3. The design of experiments is thorough and the reported results demonstrate the effectiveness of the proposed model on various graph-based tasks.

Weaknesses

1. The idea of using diffusion models is borrowed from the field of computer vision, where the diffusion and reverse process is more intuitive. However, why structured node classification would benefit from the diffusion and reverse process? Why the reverse process can be defined as a Gaussian distribution on graphs? 2. It is unclear how the proposed model is applied to inductive node classification and what benefits the diffusion model can offer in inductive settings. 3. As observed from Table 4, the performance of the proposed method is comparable or with limited improvements. More elaborations on these points are necessary. Also, the datasets used for inductive settings (except for PPI) are mostly used for transductive settings. Experiments on other relevant datasets should be added for inductive settings.

Questions

See weaknesses above.

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

No potential negative societal impact of this work.

Reviewer rLqd6/10 · confidence 2/52023-07-07

Summary

This paper proposes a new framework called the Diffusion Probabilistic Model for Structured Node Classification (DPM-SNC), which leverages diffusion probabilistic models for structured node classification on graphs. The framework aims to account for dependencies between node labels. The two key aspects of DPM-SNC are: (a) learning a joint distribution over the labels using an expressive reverse diffusion process and (b) making predictions based on known labels using manifold-constrained sampling. The authors test DPM-SNC in diverse scenarios, including partially labeled graphs in a transductive setting, as well as unlabeled graphs and in an inductive setting.

Strengths

1. The paper effectively applies probabilistic diffusion, one of the hottest techniques in the community, to the structured node classification problem. 2. DPM-SNC can be used in combination with any graph neural network (GNN) and to enhance performance on node classification tasks. This improvement is not solely empirical, as the authors provide theoretical analysis tying it to enhanced theoretical expressive power. 3. The authors conduct a comprehensive experimental evaluation, showing DPM-SNC's superior performance in both transductive and inductive node classification tasks.

Weaknesses

1. Although it is understandable that certain technical details are relegated to the appendix due to space limitations, the main paper could benefit from being more self-contained. This could be achieved by shortening the introduction or reducing some figures. Without referring to the appendix, especially for readers without expertise in diffusion models, it's difficult to grasp exactly how DPM-SNC makes predictions. 2. The paper lacks a discussion or comparison with similar works that combine deep neural networks and label propagation for structured node classification [1, 2]. These works could be used as additional baselines or in discussions related to the proposed method. 3. It appears that a limitation of the proposed method is that the diffusion process operates on (noisy versions of) labels of nodes, not their representation vectors. As a result, while DPM-SNC seems to work well for node classification problems, it's unclear whether the trained model could be effectively used as a general node embedding model for solving edge- or graph-level problems, or multi-label classification problems. [1] Huang et al., Combining Label Propagation and Simple Models Outperforms Graph Neural Networks, ICLR 2021 [2] Yoo et al., Belief Propagation Network for Hard Inductive Semi-Supervised Learning, IJCAI 2019

Questions

1. Why did the authors exclusively use GCNII for the proposed method and not for the competitor models?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

2: You are willing to defend your assessment, but it is quite likely that you did not understand the central parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

3 good

Contribution

3 good

Limitations

The authors have acknowledged the limitations of their approach.

Reviewer rLqd2023-08-14

Thank you for the response

I appreciate the authors' clear and informative reply. I anticipate that these new experimental results will be replicated in the other two datasets, Photo and Computer.

Reviewer 7A8R2023-08-14

Reply to the authors

Thank you for the detailed clarifications. I have no further comment.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC