Invited Talk
Tianxing: a Linear Complexity Transformer Model with Explicit Attention Decay for Global Weather Forecasting
Bin MU, Tongji University

Biography

Bin Mu is a Professor at the School of Computer Science and Technology, Tongji University. He leads groundbreaking interdisciplinary research at the intersection of artificial intelligence (AI), atmospheric science, and oceanography. His research portfolio encompasses cutting-edge developments in large weather models, AI-enhanced parameterization schemes, intelligent data assimilation, AI-driven post-processing for numerical weather prediction, and advanced forecasting and predictability analysis of high-impact weather and climate phenomena. Under his leadership, the AI + Atmospheric & Oceanic Sciences Research Team at Tongji University has emerged as a world-renowned research powerhouse, producing a series of seminal publications in prestigious international journals such as npj CAS and JAMES.

Beyond academic contributions, Professor Mu actively applies the latest research outcomes to real-world challenges. His AI-powered prediction systems have been adopted by internationally recognized institutions, including the International Research Institute for Climate and Society (IRI) and the Sea Ice Prediction Network (SIPN), for operational ENSO and Arctic sea ice forecasting. His expertise is consistently sought at the highest levels, with annual invitations to participate in national El Niño forecasting expert consultations, where his AI-driven insights provide critical scientific guidance for national economic planning and societal resilience strategies.

In this presentation, he will introduce the TianXing model, a transformer-based data-driven model designed with physical augmentation for skillful and efficient global weather forecasting. TianXing applies a linear complexity mechanism, with an explicit attention decay mechanism in the linear attention derived from physical insights, ensuring proportional scalability with input data size while significantly diminishing GPU resource demands. Meanwhile, a stacked autoregressive forecast strategy is employed to enhance its performance in medium-range forecasting. Notably, TianXing exhibits excellent performance, particularly in the Z500 and T850 fields, surpassing previous data-driven models and operational full-resolution models such as NCEP GFS and ECMWF IFS, as evidenced by latitude-weighted RMSE and ACC metrics. Moreover, TianXing has demonstrated remarkable capabilities in predicting extreme weather events, such as typhoons. Building on these achievements, TianXing has been invited to participate in the AI large models for weather forecasting demonstration program (AIM-FDP) launched by the China Meteorological Administration (CMA).