Approximate Cross Validation for Large Data and High Dimensions - Tamara Broderick, MIT

Explore a 35-minute lecture on approximate cross-validation techniques for large datasets and high-dimensional problems. Delve into the challenges of assessing error and variability in statistical and machine learning algorithms when dealing with modern, large-scale data. Learn about the infinitesimal jackknife (IJ) method, a linear approximation technique that can significantly speed up cross-validation and bootstrap processes. Examine finite-sample error bounds for the IJ and discover how dimensionality reduction can be applied to improve its performance in high-dimensional scenarios, particularly for leave-one-out cross-validation (LOOCV) with L1 regularization in generalized linear models. Gain insights into the theoretical foundations and practical applications of these techniques through simulated and real-data experiments, and understand how they contribute to the growing intersection of statistics and computer science in the field of machine learning.

Introduction
Crossvalidation setup
Firstorder Taylor expansion
Package
Theory
Assumptions
Experiments
Results
Conclusions