Proposal Description

Session Overview

Mixture models have become a cornerstone of data analysis, tackling tasks like clustering, anomaly detection, and modeling complex, multimodal data distributions. As technology explodes, we generate ever-increasing volumes of high-dimensional, diverse data. Traditional single-distribution models often struggle to capture this complexity.

This session brings together four renowned mixture model experts to showcase their cutting-edge research. We aim to inspire a broad audience from various research backgrounds. We'll explore the connections between theory, data, and methodological advancements. Most importantly, we'll guide potential users from diverse fields on how to leverage these advancements in their own research.

Featured Talks:

Professor Jiahua Chen: Moment Estimator and the Optimal Minimax Convergence Rate

This talk delves into the challenges of statistical inference with finite mixture models, particularly when the model order is overspecified. Professor Chen will present new findings on the moment estimator and its minimax convergence rate.

Professor Zeny Feng: Variable Selection in Mixture Regression Models

Professor Feng tackles a common challenge in mixture regression: identifying the most influential variables within each subpopulation. Her innovative algorithm optimizes regularized mixture regression models using various penalty options. This method effectively selects only the relevant covariates for each subpopulation, allowing these variables to have distinct effects on the response variable in different groups.

Professor Abbas Khalili: Feature Selection in Overfitted Finite Mixture of Regression Models

Building on recent developments in overfitted FMR models, Professor Khalili will discuss a novel penalization approach to consistently recover the true sparse structure underlying the data.

Dr. Jing Qin: Conformal Inference in Off-Policy Evaluation

This talk explores the use of mixture models in off-policy evaluation, where historical data is used to inform decision-making under a different policy. Dr. Qin will demonstrate how conformal inference can leverage mixture structures to construct efficient predictive intervals.

Target Audience:

This session is designed for researchers and practitioners with a background in statistics, machine learning, and data analysis. Attendees from diverse fields who are interested in leveraging mixture models for their research are encouraged to participate.

Expected Benefits:

By attending this session, participants will gain a deeper understanding of:

(1) Recent advancements in mixture model theory and methodology.

(2) Practical applications of mixture models to address complex data challenges.

(3) Strategies for leveraging mixture models in their own research endeavors.