Time-Series Forecasting
Where most of my work is right now.
About
I'm an undergraduate in the Automation program at Northeastern University, China. My research so far is in multivariate time-series forecasting: how channels should exchange information, and when the exogenous variables a model relies on can be trusted.
That work has led to a journal paper, a conference paper, and two patent applications. Beyond forecasting, I'm interested in large language models and embodied AI.
Research
Large Language Models
How large language models work, and how to build useful systems around them.
Embodied AI
Agents that act in the physical world, not only on text and data.
Publications
-
Conference paper
A Diagnostic Stress-Test Framework for Exogenous Variable Reliability in Transformer-based Time Series Forecasting
Projects
Controlled Local–Global Channel Communication
Multivariate long-term forecasting, where channels exchange local and global information in a controlled way.
Reliable Exogenous Variables for Time-Series Forecasting
A diagnostic stress test for how far Transformer-based forecasters can rely on their exogenous variables, plus an expert-model selection method for long-term forecasting.
Patents
Both are published Chinese patent applications, not granted patents.
-
Published application
基于受控通道通信的配电网多测点负荷预测方法及装置、存储介质、计算机设备
Informal translation, not official: Method and apparatus, storage medium, and computer device for multi-measurement-point load forecasting in distribution networks based on controlled channel communication.
-
Published application
面向外生变量长期时间序列预测的专家模型选择方法及装置
Informal translation, not official: Expert model selection method and apparatus for long-term time-series forecasting with exogenous variables.
Contact
- songyuhao799@gmail.com
- Address
- College of Information Science and Engineering,
Northeastern University, China