UTC Department of Computer Science and Engineering researchers Jingbo Wang, left, Dr. Shaogang Ren, Kegeng Tang and Dr. Zihao Wang stand before a presentation about their NeurIPS-accepted paper, “SCDM: Scalable Causal Discovery in Nonlinear Temporal Systems with Meta-Learning.” Photo by Hannah Godel.
A research team in the University of Tennessee at Chattanooga Department of Computer Science and Engineering has had a paper accepted to the main track of NeurIPS 2026, one of the world’s leading conferences in artificial intelligence and machine learning.
The paper, “SCDM: Scalable Causal Discovery in Nonlinear Temporal Systems with Meta-Learning,” was accepted to the 2026 Conference on Neural Information Processing Systems (NeurIPS). The conference is scheduled to take place in Sydney, Atlanta and Paris in December.
The UTC authors are Ph.D. students Jingbo Wang and Kegeng Tang; Dr. Zihao Wang, assistant professor and associate graduate program coordinator; and Dr. Shaogang Ren, assistant professor and graduate program coordinator.
Ren, the paper’s corresponding author, leads the research team and supervises the project. He recently received a Ruth S. Holmberg Grant for Faculty Excellence.
“We are truly excited and proud to see this work accepted by NeurIPS, one of the world’s premier conferences in artificial intelligence and machine learning, and—to the best of my knowledge—the first NeurIPS paper from UTC,” said Dr. Mengjun Xie, a Guerry Professor, UC Foundation Professor and head of the Department of Computer Science and Engineering. “This research tackles an important challenge in understanding cause-and-effect relationships in large, complex systems while significantly improving scalability and computational efficiency. It is a wonderful achievement for our faculty and students and a strong reflection of the growing excellence and visibility of AI research at UTC.”
For the 2026 cycle, NeurIPS received 30,709 valid main-track paper submissions and accepted 7,900, an acceptance rate of 25.7%.
Dr. Shaogang Ren, left, Jingbo Wang, Kegeng Tang and Dr. Zihao Wang developed SCDM by using meta-learning, an artificial intelligence approach in which a model learns across related tasks, to make the process more efficient. Photo by Hannah Godel.
According to the paper, the size of such systems complicates the work. In a system with 3,000 traffic sensors, researchers must consider nearly 9 million possible cause-and-effect links. Existing methods may require a separate model for each variable or many statistical tests, making them slow or too memory-intensive for large datasets.
The UTC team developed SCDM by using meta-learning, an artificial intelligence approach in which a model learns across related tasks, to make the process more efficient. SCDM treats prediction for each variable as a task, while sharing one prediction model and one map that measures the strength of causal relationships across the system.
Rather than processing every variable at once, the method uses small random samples in each training round. Over time, it builds a picture of the full system.
In simulated brain-imaging data based on a standard map of 100 brain regions, SCDM identified known connections more accurately than six leading methods. In a larger simulation involving about 1 million possible links, it found the correct connections more reliably than a leading recent method in the same amount of computing time.
The researchers also tested SCDM using data from 3,000 Los Angeles freeway sensors. The method ran about 23 times faster than the comparison method and used about one-fourteenth as much graphics-processing-unit memory. The strongest connections were generally between nearby sensors, a pattern expected as traffic congestion moves along freeways.
The research could support work in neuroscience and medicine by helping researchers study how brain regions influence one another in health and disease. It also could help transportation officials understand how congestion travels through road networks. The approach also could be used to study power grids, environmental monitoring, industrial equipment and financial markets, which all generate large volumes of time-series data.
The team’s next steps include enabling SCDM to select its own settings automatically and connecting it more closely with forecasting tools.
“This recognition from NeurIPS reflects the quality of research taking place at UTC,” said Dr. Kumar Yelamarthi, dean of the UTC College of Engineering and Computer Science. “Drs. Ren and Wang and their students are advancing work with the potential to help people better understand complex systems, from traffic networks to the human brain. Their achievement also shows the strength of UTC’s growing research community in artificial intelligence and machine learning.”
Learn more
College of Engineering and Computer Science
