Kaifeng Lyu 吕凯风

Research Interests

I am primarily interested in machine learning theory, AI safety/alignment, and optimization.

I believe that a good theory should come from practice and be able to advance practice: it should start from important real-world phenomena or problems, explain or solve them theoretically, and then be applied to guide the practice. Based on this philosophy, I conduct research with both theory and experiments, aiming to develop solid foundations for modern machine learning and make AI in the era of large models more efficient, safe, and reliable.

Below is a list of topics I am actively thinking about/working on:

1. Training Dynamics of Neural Networks: How do neural networks learn? How do we make the training of large models more efficient? How do training algorithms, model architectures, and training data interact with each other and affect the model's performance?
2. Modern Paradigms of Generalization in Large Foundation Models: Generalization is the ability of a model to perform well beyond its training set. Classic machine learning theory has focused more on generalization in supervised learning, but large foundation models integrate various unsupervised and supervised learning paradigms, and many new paradigms of generalization emerge. How do we understand these new generalization paradigms? How do we improve algorithms, architectures, and data to enhance the model's capabilities (such as reasoning, retrieval, in-context learning, etc.)?
3. AI Safety/Alignment: Machine learning typically optimizes a model's performance in "average cases". However, AI safety issues can manifest in extreme cases and can potentially be more catastrophic as the model becomes more intelligent. What new algorithms and paradigms do we need to mitigate or even solve the safety issues?

Recruitment for Ph.D. students

我们正在招收 2025 年秋季入学的博士生，根据情况会招录1-2个博士生。

我们希望你有钻研精神，有志于为现代机器学习方法夯实理论基础。观察到深度学习中一系列神秘的现象时，对其怀有好奇心，愿意花时间去学习和研究。我们希望你学习成绩优异，或完成过高质量的科研项目。

我们需要复合型的人才来从事这项事业。面对神经网络这类复杂系统，做好理论不仅需要善用数学工具，还要 “挽起袖子” 做实验，去观察真正的实验现象。正如开普勒发现三定律，不单需要数学，还需要从第谷的恒星数据总结规律。另一方面，光看实验而不总结规律，也会难以触及机器学习的本质。

如果你具有如下背景之一：

1. 具有扎实的数学基础，喜欢数学，对深度学习有基本的了解，愿意在博士期间做理论研究，辅以实验；
2. 具有出色的编程能力，上过深度学习相关课程或做过相关研究，愿意在博士期间基于实验发现本质，辅以理论。

欢迎感兴趣的同学与我邮件联系！请一并附上个人简历和成绩单。同时，请关注交叉信息研究院关于夏令营的信息（去年的网站）。