Graph Convolutional Networks (GCNs) have emerged as the state-of-the-art deep learning model for representation learning on graphs. However, it remains notoriously challenging to train and inference GCNs over large graph datasets, limiting their application to large real-world graphs and hindering the exploration of deeper and more sophisticated GCN graphs. This is because as the graph size grows, the sheer number of node features and the large adjacency matrix can easily explode the required memory and data movements. To tackle the aforementioned challenges, we explore the possibility of drawing lottery tickets when sparsifying GCN graphs, i.e., subgraphs that largely shrink the adjacency matrix yet are capable of achieving accuracy comparable to or even better than their full graphs. Specifically, we for the first time discover the existence of graph early-bird (GEB) tickets that emerge at the very early stage when sparsifying GCN graphs, and propose a simple yet effective detector to automatically identify the emergence of such GEB tickets. Furthermore, we advocate graph-model co-optimization and develop a generic efficient GCN early-bird training framework dubbed GEBT that can significantly boost the efficiency of GCN training by (1) drawing joint early-bird tickets between the GCN graphs and models and (2) enabling simultaneously sparsification of both the GCN graphs and models. Experiments on various GCN models and datasets consistently validate our GEB finding and the effectiveness of our GEBT, e.g., our GEBT achieves up to 80.2% ~ 85.6% and 84.6% ~ 87.5% savings of GCN training and inference costs while offering a comparable or even better accuracy as compared to state-of-the-art methods. Our source code and supplementary appendix are available at https://github.com/RICE-EIC/Early-Bird-GCN.
more »
« less
Trends in bird abundance differ among protected forests but not bird guilds
More Like this
-
-
more » « less
Abstract Various types of radar systems are increasingly being used to monitor aerial biodiversity. Each of these types has different detection capabilities and sensitivities to environmental conditions, which affect the quantity and quality of the measured objects of interest. Radar wind profilers have long been known to detect birds, but their use in ornithology has remained limited, largely because of biologists' unfamiliarity with these systems. Although the potential of radar wind profilers for quantitative bird monitoring has been illustrated with time series of raw data, a comparison with a similar radar system more established in biology is missing. Here, we compare nocturnal bird migration patterns observed by a radar wind profiler during October 2019 and April 2021 with those from a dedicated bird radar BirdScan MR1. The systems were located 50 km apart with an altitudinal difference of about 850 m. The nightly migration intensities measured with both systems were highly correlated in both spring and autumn (Pearson correlation coefficient ≈ 0.8,
P < 0.001), but estimated traffic measured by the radar wind profiler was on average five times higher in spring and nine times higher in autumn. Low ratios of the migration traffic rates of the Birdscan MR1 to those of the radar wind profiler occurred primarily in clear conditions. In both radar systems, migration occurred at significantly higher altitudes in spring than in autumn. Discrepancies in absolute numbers between both systems are likely due to both system‐inherent and external environmental and topographical factors, but also different quantification approaches. These findings support the capacity of radar wind profilers for aerial biomonitoring, independent of environmental conditions, and open up further avenues for studying the impact of weather on bird migration at detailed temporal and altitudinal scales. -
Abstract Understanding how biotic and abiotic interactions influence community assembly and composition is a fundamental goal in community ecology. Addressing this issue is particularly tractable along elevational gradients in tropical mountains that feature substantial abiotic gradients and rates of species turnover. We examined elevational patterns of avian community structure on 2 mountains in Malaysian Borneo to assess changes in the relative strength of biotic interactions and abiotic constraints. In particular, we used metrics based on (1) phylogenetic relatedness and (2) functional traits associated with both resource acquisition and tolerance of abiotic challenges to identify patterns and causes of elevational differences in community structure. High elevation communities were composed of more phylogenetically and functionally similar species than would be expected by chance. Resource acquisition traits, in particular, were clustered at high elevations, suggesting low resource and habitat diversity were important drivers of those communities. Traits typically associated with tolerance of cold temperatures and low atmospheric pressure showed no elevational patterns. All traits were neutral or overdispersed at low elevations suggesting an absence of strong abiotic filters or an increased influence of interspecific competition. However, relative bill size, which is important for thermoregulation, was larger in low elevation communities, suggesting abiotic factors were also influential there. Regardless of metric, clustered and neutral communities were more frequent than overdispersed communities overall, implying that interspecific competition among close relatives may not be a pervasive driver of elevational distribution and community structure of tropical birds. Overall, our analyses reveal that a diverse set of predominantly biotic factors underlie elevational variation in community structure on tropical mountains.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1002/eap.2377
