Curriculum Vitae
Education
- M.S. in Statistics, The University of Chicago, 2021.09 - 2023.03 (expected)
- Exchange Student, University of California - Berkeley, 2018.08 - 2018.12
- B.S. in Statistics, East China Normal University, 2015.09 - 2020.07
Research Experiences
- Test-Time Label-Shift Adaptation, 2022.06 - Present
- Supervisor: Kevin Murphy, Alexander D’Amour
- Led to the paper Beyond Invariance: Test-Time Label-Shift Adaptation for Distributions with “Spurious” Correlations. In Preparation for ICML 2023.
- Developed the EM algorithm in JAX/Flax which calculates the MAP estimator for test-time adaptation on target label distribution with neural networks and gradient boosted trees to address spurious correlation in classification
- Proposed the adoption of CXR Foundation, a pre-trained network on chest X-rays images, resulting in a 15% increase in AUC
- Synthesized spurious correlations in MNIST, CheXpert, and MIMIC-CXR datasets based on discrete copulas, and demonstrated the supremacy of our method through large-scale simulation studies on five TPU v3-8 VMs
- Feature Attribution for Gradient Boosted Trees, 2022.09 - Present
- Supervisor: Yu Wang, Bin Yu
- Led to the paper Individualized and Global Feature Attributions for Gradient Boosted Trees with l2 Regularization. Submitted to AISTATS 2023.
- Proposed a novel individualized feature attribution method for l2-regularized gradient boosted trees called Prediction Decomposition Attribution (PreDecomp) by generalizing the Saabas method
- Proposed a family of debiased global feature attribution methods called TreeInner, which achieves 10% higher AUC scores compared to state-of-the-art baselines in noisy feature identification
- Classifying Neural Networks with Type-A Quiver Representations, 2021.09 - Present
- Supervisor: Bradley Nelson, Lek-Heng Lim
- Available soon as Master’s Thesis at the University of Chicago
- Calculated the persistent barcode with LEUP factorization and shape commutation on Q and R for linear neural networks
- Conducted empirical experiments on the relationship between the barcode of neural networks and their adversarial robustness
- Analyzed the stability of real-valued barcode computation through its equivalence to the Jordan Canonical Form
- Disc Game Embedding, 2022.09-Present
- Supervisor: Alexander Strang, Lek-Heng Lim
- Led to the paper Functional Principle Trade-off Analysis: Universal Approximation via Disc Game Embedding. In Preparation for ICML 2023. Available soon on arXiv.
- Developed a novel algorithm to decompose zero-sum games into a weighted sum of disc games (continuous rock-paper-scissors games) based on a research prototype, proposing alternative algorithmic design choices along the way
- Built a front-end for the algorithm with Python and JavaScript
- CausalModel: Widely used casual inference methods, 2022.04 - 2022.06
- Action2Motion: Conditioned Generation of 3D Human Motions, 2020.01 - 2020.03
- Enhancement for Iterative Ramdon Forest, 2019.07 - 2019.10
- Supervisors: Karl Kumbier, Bin Yu
- Halved the execution time of the R package iRF for iterative random forests by rewriting critical subroutines in C++, as well as fixed bugs, improved documentation, and added test suites
- Created the R package tree.interpreter in C++/Armadillo which implements the MDI-oob algorithm for random forests
- Incorporated the MDI-oob algorithm into the iRF package, which led to higher resistance to noisy features
Working Experiences
- Maxwealth Fund, 2020.07-2021.09
- Quantitative Research Intern
Services
- AISTATS 2023 Reviewer (2022.11)
- Machine Learning Study Group Co-host at UChicago (2022.09 - Present)
- Teaching Assistant for Advanced Algebra at ECNU (Sept. 2019 - Jan. 2020)
- Teaching Assistant for Multivariate Statistics at ECNU (Feb. 2020 - Jun. 2020)
Publications
Teaching
