UBC Computer Science publishes 2 papers at top data management conference

Professor Laks Lakshmanan presented two papers and gave a workshop keynote and Professor Rachel Pottinger received a Distinguished Reviewer Award at ACM SIGMOD+PODS 2026 

UBC Computer Science debuted two papers at a top data management conference, ACM SIGMOD+PODS, which took place in Bengaluru, India from May 31- June 5, 2026. The conference is the premier forum for database researchers, practitioners, developers and users to share the latest research on data management.  

Professor Laks Lakshmanan gave a keynote lecture entitled “Toward an Open Map of the World: Can Knowledge Graphs Help?” at the Graph Data Management Experiences & Systems (GRADES) and Network Data Analytics (NDA) workshop. Most high quality maps with global or near global coverage that are available today (e.g., Google Maps and Apple Maps) are closed source. His talk will cover a machine-learning approach combined with mixed integer linear programming for map conflation, a process of integrating various open geospatial databases with high quality but local coverage to augment them into high quality maps of greater coverage, leading to the vision of an open map of the whole world. 

Professor Rachel Pottinger was also honoured as a 2026 SIGMOD Distinguished Reviewer for her excellent work in providing high-quality and timely reviews of conference papers.  

The following two papers were presented at the conference: 

1. On Efficient Approximate Aggregate Nearest Neighbor Queries over Learned Representations 
Carrie Wang, Sihem Amer-Yahia, Laks Lakshmanan, Reynold Cheng 

In this paper, researchers studied how aggregate queries, or queries that require summarizing multiple values, can be answered accurately and efficiently. Specifically, they studied Aggregation Queries over Nearest Neighbors, which compute aggregates over learned representations of the neighborhood of a query object. Answering these queries accurately and efficiently is challenging due to the high cost of generating high-quality learned representations and the variation in different aggregate functions in neighbor selection. By combining cost-effective proxy machine learning models with oracle computations, the researchers propose a query answering framework that is scalable for efficient approximation of the aggregate results. 

2. A Comprehensive Benchmark on Spectral GNNs: The Impact on Efficiency, Memory, and Effectiveness: [Experiments & Analysis] 
Ningyi Liao, Haoyu Liu, Zulun Zhu, Siqiang Luo, Laks Lakshmanan 

In this project, the researchers conducted a comprehensive study of spectral graph neural networks (GNNs), which are neural networks that use spectral methods to analyze graphs. By analyzing the efficiency, memory consumption, and effectiveness of spectral graph neural networks, the researchers provide a unified implementation  framework and a benchmark for analyzing and evaluating spectral models and filters in a fair and comprehensive manner. Specifically, their benchmark implementation allows spectral GNNs to be significantly scaled up with comparable performance and less computational cost, enabling their deployment over million-scale graphs for various tasks.