Search for: All records

Class-incremental learning (CIL) aims to continually learn a sequence of tasks, with each task consisting of a set of unique classes. Graph CIL (GCIL) follows the same setting but needs to deal with graph tasks (e.g., node classification in a graph). The key characteristic of CIL lies in the absence of task identifiers (IDs) during inference, which causes a significant challenge in separating classes from different tasks (i.e., inter-task class separation). Being able to accurately predict the task IDs can help address this issue, but it is a challenging problem. In this paper, we show theoretically that accurate task ID prediction on graph data can be achieved by a Laplacian smoothing-based graph task profiling approach, in which each graph task is modeled by a task prototype based on Laplacian smoothing over the graph. It guarantees that the task prototypes of the same graph task are nearly the same with a large smoothing step, while those of different tasks are distinct due to differences in graph structure and node attributes. Further, to avoid the catastrophic forgetting of the knowledge learned in previous graph tasks, we propose a novel graph prompting approach for GCIL which learns a small discriminative graph prompt for each task, essentially resulting in a separate classification model for each task. The prompt learning requires the training of a single graph neural network (GNN) only once on the first task, and no data replay is required thereafter, thereby obtaining a GCIL model being both replay-free and forget-free. Extensive experiments on four GCIL benchmarks show that i) our task prototype-based method can achieve 100% task ID prediction accuracy on all four datasets, ii) our GCIL model significantly outperforms state-of-the-art competing methods by at least 18% in average CIL accuracy, and iii) our model is fully free of forgetting on the four datasets. Code is available at https://github.com/mala-lab/TPP. 

Abstract Megathrusts at convergent plate boundaries generate the largest and some of the most hazardous earthquakes on Earth. However, their physical properties, including those influencing fault slip accumulation and release and earthquake‐related surface displacements, are still poorly constrained at critical depths. Here, we combine seismic imaging and geodetic modeling to investigate the structure and mechanical behavior of the Main Himalayan Thrust fault (MHT) in the center of the 2015 Mw 7.8 Gorkha rupture in Nepal. Our results from two independent observations consistently suggest the presence of a channel associated with the MHT with high compliance (shear modulus as low as ∼4 GPa) and strain anisotropy (stiffer in the vertical orientation than in the horizontal), likely arising from a weak subducting layer with north‐dipping foliation. Such mechanical heterogeneity significantly influences the quantification of short‐term fault kinematics and associated earthquake potential, with implications on across‐scale dynamics of plate boundaries in Himalaya and elsewhere. 

Abstract IntroductionAbnormal angiogenesis is central to vascular disease and cancer, and noninvasive biomarkers of vascular origin are needed to evaluate patients and therapies. Vascular endothelial growth factor receptors (VEGFRs) are often dysregulated in these diseases, making them promising biomarkers, but the need for an invasive biopsy has limited biomarker research on VEGFRs. Here, we pioneer a blood biopsy approach to quantify VEGFR plasma membrane localization on two circulating vascular proxies: circulating endothelial cells (cECs) and circulating progenitor cells (cPCs). MethodsUsing quantitative flow cytometry, we examined VEGFR expression on cECs and cPCs in four age-sex groups: peri/premenopausal females (aged < 50 years), menopausal/postmenopausal females (≥ 50 years), and younger and older males with the same age cut-off (50 years). ResultscECs in peri/premenopausal females consisted of two VEGFR populations: VEGFR-low (~ 55% of population: population medians ~ 3000 VEGFR1 and 3000 VEGFR2/cell) and VEGFR-high (~ 45%: 138,000 VEGFR1 and 39,000–236,000 VEGFR2/cell), while the menopausal/postmenopausal group only possessed the VEGFR-low cEC population; and 27% of cECs in males exhibited high plasma membrane VEGFR expression (206,000 VEGFR1 and 155,000 VEGFR2/cell). The absence of VEGFR-high cEC subpopulations in menopausal/postmenopausal females suggests that their high-VEGFR cECs are associated with menstruation and could be noninvasive proxies for studying the intersection of age-sex in angiogenesis. VEGFR1 plasma membrane localization in cPCs was detected only in menopausal/postmenopausal females, suggesting a menopause-specific regenerative mechanism. ConclusionsOverall, our quantitative, noninvasive approach targeting cECs and cPCs has provided the first insights into how sex and age influence VEGFR plasma membrane localization in vascular cells. 

The association between two event times is of scientific importance in various fields. Due to population heterogeneity, it is desirable to examine the degree to which local association depends on different characteristics of the population. Here we adopt a novel quantile-based local association measure and propose a conditional quantile association regression model to allow covariate effects on local association of two survival times. Estimating equations for the quantile association coefficients are constructed based on the relationship between this quantile association measure and the conditional copula. Asymptotic properties for the resulting estimators are rigorously derived, and induced smoothing is used to obtain the covariance matrix. Through simulations we demonstrate the good practical performance of the proposed inference procedures. An application to age-related macular degeneration (AMD) data reals interesting varying effects of the baseline AMD severity score on the local association between two AMD progression times.