Search

Graph convolutional neural networks (GCNs) are emerging neural networks for graph structures that include large features associated with each vertex. The operations of GCN can be divided into two phases - aggregation and combination. While the combination just performs matrix multiplications using trained weights and aggregated features, the aggregation phase requires graph traversal to collect features from adjacent vertices. Even though neural network applications rely on GPU’s massively parallel processing, GCN aggregation kernels exhibit rather low performance since graph processing using compressed graph structures provokes frequent irregular accesses in GPUs. In order to investigate the performance hurdles of GCN aggregation on GPU, we perform an in-depth analysis of the aggregation kernels using real GPU hardware and a cycle-accurate GPU simulator. We first analyze the characteristics of the popular graph datasets used for GCN studies. We reveal the fractions of non-zero elements in feature vectors are diverse among datasets. Based on the observation, we build two types of aggregation kernels that handle uncompressed and compressed feature vectors. Our evaluation exhibits the performance of aggregation can be significantly influenced by kernel design approaches and feature density. We also analyze the individual loads that access the data arrays of the aggregation kernels to specify critical loads. Our analysis reveals the performance of GPU memory hierarchy is influenced by access patterns and feature size of graph datasets. Based on our observations we discuss possible kernel design approaches and architectural ideas that can improve the performance of GCN aggregation.

Background: The prevalence of knee osteoarthritis(OA) in East Asia is as common for men and even higher for women than that reported in the Caucasian population. Since both population aging and economic growth have taken place at a much faster pace in Asian countries, such as South Korea, one would expect knee OA to become a major public health problem. However, few studies have examined the influence of knee OA on the quality of life (QoL) and physical function in Asia. The aim of this cross-sectional study is to investigate the influence of knee osteoarthritis (OA) on the quality of life (QoL), function and lower extremity physical performance and the gender difference in its influence in elderly community residents in Korea. Methods: Participants were from the population-based Hallym Aging Study (HAS). The mean age of the 504 study subjects was 70.2 years and 274 (54%) were women. Demographic information was obtained by questionnaire, and radiographic evaluations consisted of weight-bearing semi-flexed knee radiographs. Self-reported QoL and function were assessed using Western Ontario and McMaster Universities Osteoarthritis (WOMAC) Index and Short Form 12- item (SF-12). Performance-based lower extremity function was assessed using the tests consisting of standing balance, usual walk and chair stands. The odds ratios(ORs) for belonging to the worst quartile of WOMAC and physical performance test were calculated by logistic regression analysis in radiographic knee OA compared to non-OA after adjustment of confounders. Scores for SF-12 items were analyzed using general linear models and means adjusted for age, BMI and OA severity were compared. Results: Subjects with radiographic knee OA had significantly increased OR for belonging to the worst WOMAC quartile(for pain, 2.13,95% confidence interval[CI], 1.33-3.40, for stiffness, 2.94,95% CI,1.78-4.86, and for function, 2.97, 95% CI,1.83-4.81) and significantly worse SF-12 scores compared to non-OA after adjustment of age, BMI and sex. Women had worse WOMAC and SF-12 scores compared to men, regardless of the presence of radiographic knee OA after adjustment of age, BMI and OA severity. OA subjects had significantly worse performance score for usual walk and chair stands compared to non-OA subjects, but the ORs were no more significant after adjustment of sex. Conclusions: Knee OA negatively affects the QoL and physical function in both genders, but women are more adversely affected than men. [ABSTRACT FROM AUTHOR]