Editorial: Rising Stars In Data Mining And Management 2022

Feng Chen^1*, Shuhan Yuan^2*, Jilin Hu^3*, Cheng-Te Li^4*, Rudy Raymond^5* and Lidan Shou^6*

• ¹The University of Texas at Dallas, United States
• ²Utah State University, United States
• ³Aalborg University, Denmark
• ⁴National Cheng Kung University, Taiwan
• ⁵IBM (Japan), Japan
• ⁶Zhejiang University, China

The article by Li et al. ("Spatial data analysis for intelligent buildings: Awareness of context and data uncertainty") identifies two significant challenges in spatial data analysis for intelligent buildings. The first challenge is the complex analytical contexts of building spaces and internal entities, while the second challenge is the high spatial data uncertainty due to limitations in positioning and sensing technologies. The authors present recent advancements in addressing these challenges and propose a unified framework to accommodate modeling techniques for various analytical contexts and spatial data uncertainties. Moreover, the authors explore emerging opportunities and ongoing issues in the new technology ecosystem.The article by Rohal et al. ("AutoLoc: Autonomous Sensor Location Configuration via Cross Modal Sensing") delves into the issue of autonomous localization in floor vibration-based occupant monitoring systems. The article focuses on the applications of this technology in nonintrusive in-home continuous occupant monitoring, such as patient step tracking and gait analysis. The authors identify three primary challenges: accurate footstep event prediction and localization via video signal with noisy posture estimation, solving multilateration equations with unknown vibration propagation velocity, and footstep event selection from accumulating sensor data to achieve precise vibration sensor localization. The authors propose a scheme combining physical and data-driven knowledge to overcome these challenges. Knowledge-infused learning has attracted tremendous attention thanks to the improved knowledge graphs that can represent meaningful relations between entities at a large scale. The paper entitled "Knowledge-infused learning for entity prediction in driving scenes" by Wickramarachchi et al. addresses the challenge of scene understanding within the autonomous driving domain. The authors leverage heterogeneous, high-level semantic knowledge graphs of driving scenes to predict potentially unrecognized entities and improve driving-scene understanding. They developed an innovative neuro-symbolic solution that utilizes an expressive, holistic representation of the scene with knowledge graphs and conducts entity prediction based on knowledge-graph embeddings.In the era of big data, graph structures are widely used to represent complex relational information. Therefore, various graph-based algorithms and tools have been developed to address different real-world tasks, such as recommendation, fraud detection, and molecule design. The article by Fu et. al. ("Natural and Artificial Dynamics in Graphs: Concept, Progress, and Future") provides a comprehensive review of graph-based algorithms and tools with a focus on the topic of natural dynamics and artificial dynamics in graphs. Natural dynamics refers to the evolving topology structures, node-level, edge-level, and subgraph-level features and labels of input graphs, while artificial dynamics involves changes made by end-users to existing or nonexisting graph-related elements to improve performance. In this article, the authors first introduce three topics in graph research, namely graph mining, graph representations, and graph neural networks (GNNs). Then, the authors present the definitions of natural dynamics and artificial dynamics in graphs and their related studies as well as discuss the interplay between natural and artificial dynamics and their impact on graph research topics.Dynamic transfer learning indicates transferring knowledge from a static source task with adequate label information to a dynamic target task with little or no label information. Most existing theoretical studies and practical algorithms of dynamic transfer learning assume that the target task is continuously evolving over time, which is not always true in real-world scenarios, e.g., the target distribution is suddenly changing at some time stamp. To address this limitation, the article by Wu et. al. (Dynamic transfer learning with progressive meta-task scheduler) develops a meta-learning framework, denoted as L2S, for dynamic transfer learning based on a progressive meta-task scheduler. The key idea is to incrementally learn to schedule the meta-pairs of tasks and then learn the optimal model initialization from those meta-pairs of tasks for fast adaptation to the newest target task. The effectiveness of the L2S framework is verified both theoretically and empirically.We hope these papers will inspire the scientific community to conduct research across the entire breadth of Data Mining and Management. We are grateful to the authors and referees for their tremendous contributions and efforts toward making this research topic possible.