CurvDrop: A Ricci Curvature Based Approach to Prevent Graph Neural Networks from Over-Smoothing and Over-Squashing

Liu, Yang; Zhou, Chuan; Pan, Shirui; Wu, Jia; Li, Zhao; Chen, Hongyang; Zhang, Peng

doi:10.1145/3543507.3583269

Cited by 14 publications

(3 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Trade-offs between over-smoothing and oversquashing. In recently years, an increasing number of works based on graph rewiring focus on handling both over-smoothing and over-squashing [247,248,250,256] and the tradeoffs between between over-smoothing and over-squashing is emphasized. Intuitively, the graph rewiring methods to alleviate over-squashing often introduce additional edges to original graph, which exerts a smoothing effect on the graph thus poses a risk of over-smoothing.…”

Section: Discussionmentioning

confidence: 99%

“…The above discussion highlights the importance of amplifying information flow from distant nodes to alleviate over-squashing in graphs and the aforementioned graph rewiring is an effective to achieve the goal. There are wo main types of graph rewiring methods that are spatial graph rewiring [221,244,247,248,249,250,251,252] and spectral graph rewiring [246,253,254,255,256,257,258]. The spatial graph rewiring usually connect one node to another node within its receptive field while spectral graph rewiring often optimize some metrics that measures connectivity of graphs.…”

Section: Over-squashingmentioning

confidence: 99%

See 1 more Smart Citation

The prevalence and characteristics of metabolic syndrome according to different definitions in China: a nationwide cross-sectional study, 2012–2015

et al. 2022

View full text Add to dashboard Cite

Background Metabolic syndrome (MetS) is characterized by a cluster of signs of metabolic disturbance and has caused a huge burden on the health system. The study aims to explore the prevalence and characteristics of MetS defined by different criteria in the Chinese population. Methods Using the data of the China Hypertension Survey (CHS), a nationally representative cross-sectional study from October 2012 to December 2015, a total of 28,717 participants aged 35 years and above were included in the analysis. The MetS definitions of the International Diabetes Federation (IDF), the updated US National Cholesterol Education Program Adult Treatment Panel III (the revised ATP III), and the Joint Committee for Developing Chinese Guidelines (JCDCG) on Prevention and Treatment of Dyslipidemia in Adults were used. Multivariable logistic regression was used to identify factors associated with MetS. Results The prevalence of MetS diagnosed according to the definitions of IDF, the revised ATP III, and JCCDS was 26.4%, 32.3%, and 21.5%, respectively. The MetS prevalence in men was lower than in women by IDF definition (22.2% vs. 30.3%) and by the revised ATP III definition (29.2% vs. 35.4%), but the opposite was true by JCDCG (24.4%vs 18.5%) definition. The consistency between the three definitions for men and the revised ATP III definition and IDF definition for women was relatively good, with kappa values ranging from 0.77 to 0.89, but the consistency between the JCDCG definition and IDF definition (kappa = 0.58) and revised ATP III definition (kappa = 0.58) was poor. Multivariable logistic regression showed that although the impact and correlation intensity varied with gender and definition, area, age, education, smoking, alcohol use, and family history of cardiovascular disease were factors related to MetS. Conclusions The prevalence and characteristics of the MetS vary with the definition used in the Chinese population. The three MetS definitions are more consistent in men but relatively poor in women. On the other hand, even if estimated according to the definition of the lowest prevalence, MetS is common in China.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Over-squashingmentioning

confidence: 99%

The prevalence and characteristics of metabolic syndrome according to different definitions in China: a nationwide cross-sectional study, 2012–2015

et al. 2022

View full text Add to dashboard Cite

show abstract

“…In practice, the required number of graph layers grows with the multi-hop dependency between two distant nodes (Wang et al, 2021;Hong et al, 2022), which inevitably encounters the problem of oversmoothing . Some researchers have conducted studies on relieving this problem Min et al, 2020;Huang and Li, 2023;Yang et al, 2023;Liu et al, 2023;Koishekenov, 2023;Song et al, 2023). However, these methods are all breadth-aggregation-based, i.e., they only posed adjustments on breadth aggregation like improving the aggregation filters, scattering the aggregation target using the probabilistic tool, etc., but never jumping out it.…”

Section: Related Workmentioning

confidence: 99%

Identity of university Chinese heritage language learners in Hong Kong

Li¹,

李蓁²

View full text Add to dashboard Cite

The core of out-of-distribution (OOD) detection is to learn the in-distribution (ID) representation, which is distinguishable from OOD samples. Previous work applied recognition-based methods to learn the ID features, which tend to learn shortcuts instead of comprehensive representations. In this work, we find surprisingly that simply using reconstruction-based methods could boost the performance of OOD detection significantly. We deeply explore the main contributors of OOD detection and find that reconstructionbased pretext tasks have the potential to provide a generally applicable and efficacious prior, which benefits the model in learning intrinsic data distributions of the ID dataset. Specifically, we take Masked Image Modeling as a pretext task for our OOD detection framework (MOOD). Without bells and whistles, MOOD outperforms previous SOTA of one-class OOD detection by 5.7%, multi-class OOD detection by 3.0%, and near-distribution OOD detection by 2.1%. It even defeats the 10-shot-per-class outlier exposure OOD detection, although we do not include any OOD samples for our detection.

show abstract

Insight at the right spot: Provide decisive subgraph information to Graph LLM with reinforcement learning

Shen,

Cambria,

Wang

et al. 2025

Information Fusion

View full text Add to dashboard Cite

CurvDrop: A Ricci Curvature Based Approach to Prevent Graph Neural Networks from Over-Smoothing and Over-Squashing

Cited by 14 publications

References 16 publications

The prevalence and characteristics of metabolic syndrome according to different definitions in China: a nationwide cross-sectional study, 2012–2015

The prevalence and characteristics of metabolic syndrome according to different definitions in China: a nationwide cross-sectional study, 2012–2015

Identity of university Chinese heritage language learners in Hong Kong

Insight at the right spot: Provide decisive subgraph information to Graph LLM with reinforcement learning

Contact Info

Product

Resources

About