I have participated in many scientific projects in companies and schools, such as Few-Shot Segmentation and Self-Supervised Visual Representation. The main projects are as follows:

	Weather Forcasting (2023.06-2024.01) GNN: We try to build a global weather forcasting system with AI model, which seems the global weather signals as a Latitude and longitude graph. We design a GCN model to forcast the global weather. SKNO: We design a new SKNO model, which integrates the KNO model and SHT operator. SKNO is developed based on the 16-year assimilation reanalysis dataset ERA5 and can forecast global weather with 6-hour temporal resolution and 1.4-degree spatial resolution, including temperature, humidity, wind and other indicators. The related research has been pushed in [code].
	Self-Prompt Learning (2023.04-2023.10) SPPEL : Different from the general prompt learning with extra-prompt support, we try to design a model with the prompt support from the input of it. The related work has been accepted by TCSVT.
	Few-Shot Segmentation (2022.04-2023.06) MCINet : We design a new model to fuse different contents from learnable branch and frozen backbone and achieve SOTA in 8 standard benchmarks for PASCAL and COCO datasets with different backbones (ResNet50 and ResNet101). The related research has been received by TOMM. DAM and PMNet : Further explore lightweight few-shot segmentation models. With a parameter of 0.68M, the model shows good performance on Cross-class, Cross-dataset and Cross-domain tasks. The related research has been accepted by WACV2024 and Neucom.
	Self-supervised Visual Representation (2021.11-2022.06) Self Supervised Learning: We design a model with higher accuracy and lower computation to encode visual signal. Finally, when GFLOPS were reduced to half of the vit-base, the model performance did not significantly degrade.