Jiaye Teng

Research Field

The ultimate goal of the research group is to understand neural networks theoretically and open the "black box" of deep learning. Specifically, we hope to ultimately achieve: through theory

1. Explain the existing model phenomena,

2. For a given dataset, model, and algorithm, be able to predict model outcomes without training.

3. Able to find corresponding effective (or even optimal) models and algorithms for a given data structure.

The small questions that I am currently interested in include:

1. Generalization-related topics, including but not limited to the use of algorithm stability in non-convex optimization, implicit regularization of special model structures, the enhancement of generalization theory through trajectory properties, the interaction between network structure data formats and algorithms, etc.

2. Large model related, including but not limited to how to accelerate large models, how to compress large models, the uncertainty issues of large models, etc.

3. Conformal prediction-related, including but not limited to the relaxation of conformal prediction assumptions, modeling of different data formats, etc.

4. Causal inference-related, including but not limited to causal inference of nonlinear problems, robustness of causal inference, etc.

Of course, if you have other interesting questions, feel free to contact me and introduce me to them.

Education Background

September 2020 - July 2024 Tsinghua University Institute for Interdisciplinary Information Research PhD

September 2016 - July 2020 Shanghai University of Finance and Economics School of Statistics and Management Bachelor's Degree

Work Experience

July 2024--Present, Shanghai University of Finance and Economics, School of Statistics and Management, Assistant Professor.