/ research
Neural operators, medical-device geometry optimization, and real-time small-object detection, with papers, code, and DOI links attached.
Tianle Xu
Introduces MPNO, a mixture-of-experts physics-informed neural operator that uses multiple Fourier-operator expert heads to model multi-branch steady-state Bratu solution sets, separating branches on 31/32 Bratu-1D parameter values and validating residual training on Bratu-2D.
MPNO Project Draft Manuscript
Xin Huang, Tianle Xu, Chang Jun Ren, Yuhuang Chen, Jia Shijie
Introduces ARRT-DETR, a real-time small-object detector that replaces standard additive decoder residuals in RT-DETR with attention-based cross-layer query aggregation. On VisDrone2019, the best three-layer configuration improves RT-DETR-L by 7.2 mAP50 and 5.2 mAP50-95 while reducing latency by 1.3 ms.
BMVC 2026 Conference Submission
Tianle Xu, Zihan Zhu, Yunhan Cai, Shunping Chen, Jia Guo, Shengzhang Wang
Published in Bio-Design and Manufacturing. Presents a design-simulation-manufacturing-assessment workflow for durable polymeric aortic valves, combining B-spline leaflet parameterization, FEM, and NSGA-II multi-objective optimization, then validating the optimized dip-molded prototype through pulsatile-flow and accelerated-wear tests.
Bio-Design and Manufacturing 8, 835-846