Robust learning
Generalization under distribution shift and reliable evaluation.
Research · Engineering · Notes
Learning systems that stay useful beyond the lab.
I explore robust machine learning and turn research ideas into dependable, reproducible software.
About
Evidence first, systems thinking throughout.
My work sits between machine learning research and practical engineering: careful experiments, clear explanations, and systems that can be operated with confidence.
Research
Current interests
Representation learning
Useful structure, efficient adaptation, and interpretable evidence.
Research systems
Reproducible pipelines that connect experiments with production.
Projects
Selected work
Domain Generalization Research
activeOngoing study of robust representation learning and model behavior under distribution shift, with an emphasis on reproducible experiments.
Ximeng Site Platform
activeA reproducible static publishing platform that combines an Astro profile, a Hugo knowledge base, and atomic server deployments.
Blog
Latest posts
- 1.1 时间序列预测到底在预测什么？
从信息边界、预测目标与损失函数出发，建立时间序列预测的基本问题定义。
- 1.3 多步预测策略：Recursive、Direct 与 Multi-Output
比较递推、直接预测和多输出策略，理解误差传播、训练成本与输出依赖的取舍。
- 1.2 Lookback、Horizon 与 Sliding Window
从预测时刻出发构造滑动窗口，理解回看长度、预测长度和步长对样本及评估的影响。
Connect