UBC Computer Science shines at top international data mining conference
Researchers presented new findings at ACM SIGKDD, including a new framework for fair resource allocation and a new foundation for developing and evaluating intrusion detection systems,
From census results to healthcare records to student class records, there are vast amounts of data about different aspects of our lives. In order to make sense of the information, researchers must analyze and find patterns in data, often using computational tools and techniques to automate the process and tackle large datasets.
To share the latest knowledge in the field, UBC researchers were at the 2026 ACM Special Interest Group on Knowledge Discovery and Data Mining (SIGKDD) from August 9-13, 2026 in Jeju, Korea. SIGKDD is one of the leading international conferences for researchers to share the latest in data mining as well as data science and machine learning.
One of the papers, led by former UBC postdoctoral fellow and current professor at the School of Computer Science and Technology at Huazhong University of Science and Technology, Keke Huang, was selected for an oral presentation, representing the top 20% of accepted papers at the conference.
The following papers were presented at the conference:
DECKBench: Benchmarking Multi-Agent Frameworks for Academic Slide Generation and Editing
Daesik Jang, Morgan Lindsay Heisler, Linzi Xing, Yifei Li, Edward Wang, Ying Xiong, Yong Zhang, Zhenan Fan
In this paper, researchers built Deck Edits and Compliance Kit Benchmark (DECKBench), an evaluation framework for academic slide generation and editing. DECKBench assesses the entire slide deck workflow, from generating slides based on a research paper to iterative editing of the slide deck based on user instructions. The results of the paper underscore the need for standardized and reproducible evaluation of academic slide generation and editing.
Parameterized Fair Resource Allocation under Diversity Constraints
Keke Huang, Yik Yu Ng, Laks V.S. Lakshmanan, Xiaokui Xiao
*Oral presentation
From housing assignments to government auctions to school course assignments, the problem of resource allocation exists in many domains. In this paper, researchers present PRA, a parameterized framework for resource allocation under diversity constraints. The researchers found that PRA can regulate group-level diversity and yield optimally fair assignments within diversity constraints, overall outperforming existing baselines on effectiveness and robustness in real-world applications.
PIDSMaker: Building and Evaluating Provenance-based Intrusion Detection Systems
Tristan Bilot, Baoxiang Jiang, Thomas Pasquier
In this paper, researchers present PIDSMaker, an open-source framework for developing and evaluation provenance-based intrusion detection systems (PIDS), which are applications that detect stealthy cybersecurity threats. PIDSMaker consolidates eight systems into a modular architecture to allow for consistent evaluation and fair comparison. Moreover, it includes a YAML-based configuration interface that allows researchers to assemble new PIDS efficiently.
SOP-Bench: Complex Industrial SOPs for Evaluating LLM Agents
Subhrangshu Nandi, Arghya Datta, Rohith Nama, Udita Patel, Nikhil Vichare, Indranil Bhattacharya, Shivam Asija, Arushi Gupta, Giuseppe Carenini, Jing Xu, Shayan Ray, Huzefa Raja, Aaron Chan, Francesco Carbone, Esther Xu Fei, Gaoyuan Du, Zuhaib Akhtar, Prince Grover, Sreyoshi Bhaduri, Weian Chen, Wei Zhang, Ming Xiong, Harshita Asnani, Jeetu Mirchandani
In this paper, researchers present SOP-Bench, a benchmark for evaluating LLM-based agents carrying out complicated standard operation procedures (SOPs). SOP-Bench includes over 2000 executable tasks from SOPs in diverse domains, including health care, customer service and financial compliance. The researcher’s baseline experiments across representative frontier models confirm their finding that agent performance depends on the task and context, making systematic evaluation on the realistic SOPs of SOP-Bench crucial before the agent can be released to production.