Stat 319: Literature of Statistics (data fusion, meta-analysis, transportability and all that)

Overview

There are lots of settings in applied statistics where we must use multiple data sets to solve one problem. There are many ways that this comes up and many proposed solutions with partial but not complete overlap. The work has been done under terms such as data fusion, meta-analysis, transportability, domain adaptation, distribution shift, hierarchical models, statistical matching, record linkage and more. There are also bespoke solutions to very specific problems that may not necessarily define an umbrella term. It may even seem that most serious applied problems require one to use multiple data sets and then it seems odd that most statistics courses are about how to work with just one data set.

This literature course is about exploring and explaining as many of those methods as we can.

Instructor

Art Owen
Sequoia Hall 130

Schedule of topics

Here are the slides that students presented. In a few cases they opted to provide an article instead.
Date Topic 1 Topic 2
April 3 Ivy Zhang on Embeddings Michael Salerno on prediction powered inference
April 10 Joon Lee on shrinking \(\hat\beta_{\text{Target data}}\) towards \(\hat\beta_{\text{Large data}}\) A paper on pooling surveys
April 17 Henry Smith on Meta-analysis under dependence Ginnie Ma on Partial conjunction testing
April 24 Aaditya Ghosh on Distribution shift (following Rothenhausler++) Valerie Ho on Dueling Bayesians
May 1 Viet Vu on Distribution shift (following Sur++) Yanxin Zhou on combining data in astrophysics
May 8 Noah Cowan on Distributionally robust optimization Allison Xu on satellite data vs ground truth
May 15 Jack Krew on transfer learning A paper on multi-omics
May 22 Etaash Katiyar on Ecological inference Dileka Gunawardana on Proxy data
May 29 Zitong Yang on AI for math olympiad problems using web data and LLM reasoning Sarah Zhao on synthetic data and GWAS
June 5 Ziang Song on federated learning Meta-analysis and Bayesian hacking from Data Colada
June 5 Wenlong Li on learning to learn Yash Nair on classic p-value meta-analysis

Evaluation

Evaluation is based on presentations, written slides, and class participation.