Brief Introduction

☺️I am a third-year undergraduate student majoring in Software Engineering at Harbin Institute of Technology. My research interests include continual learning, video and image processing, image generation, and multimodal learning. At my university, I have participated in research at the ICES Lab under the guidance of Prof. Chunshan Li. Additionally, I have interned at Westlake University advised by Prof. Ziqing Li, and at the OV³Lab at Peking University advised by Prof. Jiahuan Zhou.

📚 Research Experience

🌈 Deep Priors Driven Posterior Sampling in Diffusion Models

Duration: 2024.03–2024.06
Role: Core Member

Proposed a deep prior-driven method to address the linear inverse problem in diffusion models.
Improved image restoration capability by guiding the diffusion process with deep neural networks.
Conducted denoising, inpainting, and colorization experiments on the FFHQ face dataset, using PSNR and LPIPS as evaluation metrics, and achieved better performance compared to previous works.

📹 National College Student Innovation Training Project — Remote Sensing Meteorological Video Prediction

Duration: 2024.03–2024.06
Role: Core Member

Utilized spatiotemporal sequence prediction models to address the problem of accurately predicting cloud changes and weather conditions.
Extracted spatiotemporal features from cloud video data to learn dynamic movement patterns of cloud layers.

🧪 Prototype-based Dual-Dimensional Knowledge Distillation for Online Continual Learning

Duration: 2024.11–2025.02
Role: First Author

Proposed an optimization method combining prototype-based dual-dimensional knowledge distillation to balance model learning ability and forgetting resistance in online continual learning.
Conducted classification experiments on standard datasets like CIFAR-100, demonstrating superior performance in terms of average accuracy and forgetting rate compared to existing baseline models.
📑The paper is currently under review for a CCF-C conference.

💼 Research Internships

Westlake University Prof. Li Ziqing Image Reconstruction

Applied the reconstruction module to image generation and multimodal fields, enhancing image reconstruction capability by quantizing vectors and aligning semantic information from multiple modalities in the latent space.
Contributed to literature review and paper reproduction, testing and evaluating the performance of the reconstruction algorithm, and integrating this algorithm into the feature alignment module of a multimodal model.

Peking University Wang Xuan Institute Prof. Zhou Jiahuan Online Continual Learning

Introduced multi-level prototypes to reduce the model’s forgetting rate from both inter-task and intra-task perspectives.
📑The paper is currently under review for TPAMI (CCF-A).
Contributed to early-stage research, expanding the multi-level prototype method, implementing it on the TinyImageNet dataset, and writing the experimental sections of the paper.

📱 Software Engineering Projects

As a Software Engineering student, I have developed solid software development skills. I am proficient in using frameworks such as Vue for front-end development and Spring Boot for back-end services, enabling me to create robust and efficient full-stack applications. Below are some of the software engineering projects I have worked on:

🌟 Research Passion and Future Goals

I firmly believe that “learning is a never-ending journey,” which serves as my driving force in the pursuit of research. In my future research endeavors, I will continue to embrace lifelong learning, maintain curiosity about new technologies and methods, and accumulate more knowledge and experience through continuous literature review and practice. My goal is to apply theoretical knowledge to real-world problems and continuously enhance my abilities in the fields of computer vision and machine learning.

Xiaochen Yang