About

我是 Leo Xia，西安交通大学人工智能专业本科生。目前正在罗敏楠老师组内收尾一个足球长视频理解项目，同时也在探索其他方向，对空间智能、音频和具身智能尤其感兴趣。我常常思考如何将模型架构创新与这些方向的结合，比如应用 Energy Transformer、Nueral Symbolic 等思路，尝试在保留模型能力的同时，降低训练和推理成本，提高实时性。

我喜欢从代码入手：复现方法、跑实验，再从结果中寻找新的问题。对于偏底层架构创新的工作，也会深入钻研背后的数学原理。这个博客记录了我的学习笔记与实验过程，欢迎对相关问题感兴趣的老师和同学交流。

I'm Leo Xia, an undergraduate studying Artificial Intelligence at Xi’an Jiaotong University. I'm currently wrapping up a project on long-form soccer video understanding in Professor Luo’s research group. I often think about how to combine innovative model architectures with these approaches—such as applying concepts like the Energy Transformer and Neural Symbolic—to reduce training and inference costs and improve real-time performance while preserving the model’s capabilities.

I like to start with code: reproduce methods, run experiments, and let the results lead to new questions. When it comes to modifications in low-level architecture or methodology, I also break down the underlying mathematical principles. This blog is where I keep my learning notes and experimental records. I'm always happy to exchange ideas with researchers and fellow students who share these interests.