<ul data-eligibleForWebStory="true"><li>Graph-based Active Learning (AL) uses graph structure to prioritize label queries efficiently, reducing costs and user burden in dynamic sensing environments.</li><li>Existing graph AL methods are typically evaluated on static graph datasets, focusing on prediction accuracy rather than user-centric considerations like sampling diversity, query fairness, and adaptability to dynamic settings.</li><li>To address this gap, GRAIL is introduced as a benchmarking framework to evaluate graph AL strategies in dynamic, real-world environments.</li><li>GRAIL introduces new metrics to assess sustained effectiveness, diversity, and user burden, offering a comprehensive evaluation of AL methods in varying conditions.</li><li>Experiments on datasets with real-life human sensor data demonstrate trade-offs between prediction performance and user burden, revealing limitations in current AL strategies.</li><li>GRAIL emphasizes the importance of balancing node importance, query diversity, and network topology for evaluating graph AL solutions in dynamic environments.</li></ul>

GRAIL: A Benchmark for GRaph ActIve Learning in Dynamic Sensing Environments

Discover more