Yuekai Sun

I am an associate professor of statistics at the University of Michigan. My research leverages statistical science to make AI more safe and reliable in the real world. Some topics of recent interest include

AI evaluation & safety,
algorithmic fairness,
learning under distribution shifts.

Before coming to Michigan, I obtained my PhD in computational mathematics from Stanford University, where I worked with Michael Saunders and Jonathan Taylor, and my BA (also in computational math) from Rice University.

Publications

Sloth: scaling laws for LLM skills to predict multi-benchmark performance across families

Felipe Maia Polo, Seamus Somerstep, Leshem Choshen, Yuekai Sun, M. Yurochkin

ABS HTML PDF

Distributionally Robust Performative Prediction

Songkai Xue, Yuekai Sun

ABS HTML PDF

Microfoundation Inference for Strategic Prediction

Daniele Bracale, Subha Maity, Felipe Maia Polo, Seamus Somerstep, Moulinath Banerjee, Yuekai Sun

ABS HTML PDF

LiveXiv - A Multi-Modal Live Benchmark Based on Arxiv Papers Content

Nimrod Shabtay, Felipe Maia Polo, Sivan Doveh, Wei Lin, M. Mirza, Leshem Choshen, M. Yurochkin, Yuekai Sun, Assaf Arbelle, Leonid Karlinsky, Raja Giryes

arXiv.org 2024

ABS HTML PDF

Algorithmic Fairness in Performative Policy Learning: Escaping the Impossibility of Group Fairness

Seamus Somerstep, Ya'acov Ritov, Yuekai Sun

Conference on Fairness, Accountability and Transparency 2024

ABS HTML PDF

Efficient multi-prompt evaluation of LLMs

Felipe Maia Polo, Ronald Xu, Lucas Weber, M'irian Silva, Onkar Bhardwaj, Leshem Choshen, Allysson Flavio Melo de Oliveira, Yuekai Sun, M. Yurochkin

arXiv.org 2024

ABS HTML PDF

A transfer learning framework for weak-to-strong generalization

Seamus Somerstep, Felipe Maia Polo, Moulinath Banerjee, Ya'acov Ritov, M. Yurochkin, Yuekai Sun

ABS HTML PDF

Learning the Distribution Map in Reverse Causal Performative Prediction

Daniele Bracale, Subha Maity, Moulinath Banerjee, Yuekai Sun

arXiv.org 2024

ABS HTML PDF

Prompt Exploration with Prompt Regression

Michael Feffer, Ronald Xu, Yuekai Sun, M. Yurochkin

arXiv.org 2024

ABS HTML PDF

Learning In Reverse Causal Strategic Environments With Ramifications on Two Sided Markets

Seamus Somerstep, Yuekai Sun, Y. Ritov

International Conference on Learning Representations 2024

ABS HTML PDF

Aligners: Decoupling LLMs and Alignment

Lilian Ngweta, Mayank Agarwal, Subha Maity, Alex Gittens, Yuekai Sun, M. Yurochkin

Tiny Papers @ ICLR 2024

ABS HTML PDF

tinyBenchmarks: evaluating LLMs with fewer examples

Felipe Maia Polo, Lucas Weber, Leshem Choshen, Yuekai Sun, Gongjun Xu, M. Yurochkin

International Conference on Machine Learning 2024

ABS HTML PDF

Weak Supervision Performance Evaluation via Partial Identification

Felipe Maia Polo, Subha Maity, M. Yurochkin, Moulinath Banerjee, Yuekai Sun

ABS HTML PDF

Fusing Models with Complementary Expertise

Hongyi Wang, Felipe Maia Polo, Yuekai Sun, Souvik Kundu, Eric P. Xing, M. Yurochkin

International Conference on Learning Representations 2023

ABS HTML PDF

An Investigation of Representation and Allocation Harms in Contrastive Learning

Subha Maity, Mayank Agarwal, M. Yurochkin, Yuekai Sun

International Conference on Learning Representations 2023

ABS HTML PDF

Large Language Model Routing with Benchmark Datasets

Tal Shnitzer, Anthony Ou, M'irian Silva, Kate Soule, Yuekai Sun, Justin Solomon, Neil Thompson, M. Yurochkin

arXiv.org 2023

Yuekai Sun

Publications

Sloth: scaling laws for LLM skills to predict multi-benchmark performance across families

Distributionally Robust Performative Prediction

Microfoundation Inference for Strategic Prediction

LiveXiv - A Multi-Modal Live Benchmark Based on Arxiv Papers Content

Algorithmic Fairness in Performative Policy Learning: Escaping the Impossibility of Group Fairness

Efficient multi-prompt evaluation of LLMs

A transfer learning framework for weak-to-strong generalization

Learning the Distribution Map in Reverse Causal Performative Prediction

Prompt Exploration with Prompt Regression

Learning In Reverse Causal Strategic Environments With Ramifications on Two Sided Markets

Aligners: Decoupling LLMs and Alignment

tinyBenchmarks: evaluating LLMs with fewer examples

Weak Supervision Performance Evaluation via Partial Identification

Fusing Models with Complementary Expertise

An Investigation of Representation and Allocation Harms in Contrastive Learning

Large Language Model Routing with Benchmark Datasets

Conditional independence testing under misspecified inductive biases

ISAAC Newton: Input-based Approximate Curvature for Newton's Method

On Uniform Consistency of Spectral Embeddings

Selective Degradation of Hemicellulose and Lignin for Improving Enzymolysis Efficiency via Pretreatment Using Deep Eutectic Solvents.

Simple Disentanglement of Style and Content in Visual Representations

Calibrated Data-Dependent Constraints with Exact Satisfaction Guarantees

How does overparametrization affect performance on minority groups?

Understanding new tasks through the lens of training data via exponential tilting

Predictor-corrector algorithms for stochastic optimization under gradual distribution shift

Domain Adaptation meets Individual Fairness. And they get along

Achieving Representative Data via Convex Hull Feasibility Sampling Algorithms

A linear adjustment-based approach to posterior drift in transfer learning.

Post-processing for Individual Fairness

On sensitivity of meta-learning to support data

Individually Fair Gradient Boosting

Statistical inference for individual fairness

Individually Fair Ranking

Outlier-Robust Optimal Transport

Matrix completion methods for the total electron content video reconstruction

There is no trade-off: enforcing fairness can improve accuracy

Does enforcing fairness mitigate biases caused by subpopulation shift?

SenSeI: Sensitive Set Invariance for Enforcing Individual Fairness

Two Simple Ways to Learn Individual Fairness Metrics from Data

Minimax optimal approaches to the label shift problem in non-parametric settings

Auditing ML Models for Individual Bias and Unfairness

Federated Learning with Matched Averaging

Communication-Efficient Integrative Regression in High-Dimensions

Meta-analysis of heterogeneous data: integrative sparse regression in high-dimensions

Learning fair predictors with Sensitive Subspace Robustness

Training individually fair ML models with sensitive subspace robustness

Dirichlet Simplex Nest and Geometric Inference

Uniform Bounds for Invariant Subspace Perturbations

Precision Matrix Estimation with Noisy and Missing Data

Creation and analysis of biochemical constraint-based models using the COBRA Toolbox v.3.0

Statistical convergence of the EM algorithm on Gaussian mixture models

Debiasing representations by removing unwanted variation due to protected attributes

Creation and analysis of biochemical constraint-based models: the COBRA Toolbox v3.0.

Supplemental - A Geometric Approach to Archetypal Analysis and Nonnegative Matrix Factorization

An inexact subsampled proximal Newton-type method for large-scale machine learning

On conditional parity as a notion of non-discrimination in machine learning

Evaluating the statistical significance of biclusters

Do genome-scale models need exact solvers or clearer standards?

Systems biology definition of the core proteome of metabolism and expression is consistent with high-throughput data

Communication-efficient sparse regression: a one-shot approach

A Geometric Approach to Archetypal Analysis and Nonnegative Matrix Factorization

Random projections for non-negative matrix factorization

Valid post-correction inference for censored regression problems

Exact post-selection inference, with application to the lasso

Exact inference after model selection via the Lasso

Exact post-selection inference with the lasso

Modulation of the work function of fullerenes C60 and C70 by alkali-metal adsorption: A theoretical study

Learning Mixtures of Linear Classifiers

Robust flux balance analysis of multiscale biochemical reaction networks

On model selection consistency of regularized M-estimators

On model selection consistency of M-estimators with geometrically decomposable penalties

Proximal Newton-Type Methods for Minimizing Composite Functions

Proximal Newton-type Methods for Minimizing Convex Objective Functions in Composite Form

Proximal Newton-type methods for convex optimization

Nanostructure on taro leaves resists fouling by colloids and bacteria under submerged conditions.

Regular and reverse nanoscale stick-slip behavior: Modeling and experiments

Nanoscale Friction Dynamic Modeling

Synchrotron radiation photoemission spectroscopic study of band offsets and interface self-cleaning by atomic layer deposited HfO2 on In0.53Ga0.47As and In0.52Al0.48As