Repositories for papers

The papers can also be found in clarte_phd.pdf at Chapters 1 to 7 (except for Chapter 6). They are the following :

• Theoretical characterization of uncertainty in high-dimensional linear classification : Link to paper, Github repo

• On double-descent in uncertainty quantification in overparametrized models : Link to paper, Github repo

• Expectation consistency for calibration of neural networks : Link to paper, Github repo

• Analysis of Bootstrap and Subsampling in High-dimensional Regularized Regression : Link to paper, Github repo

• Building Conformal Prediction Intervals with Approximate Message Passing : Link to paper, Github repo

Remarks on the code The codes are written in Python, Julia or Rust. Rust code requires compilation into a Python package, the instructions are provided in this repository.

Work on ensembling

• The folder ensembling contains : codes and reports regarding ensembling in the context of random feature models, see Chapter 6 of Lucas Clarte's manuscript in clarte_phd.pdf. The files ensembling_files.zip and ensembling_output.pdf contains the notes on this model. The sub-folder random-features contains experiments on ensembling random features.

Work on mean-variance estimators

• Unpublished work on mean-variance estimators : Github repo. The folder mean-variance contains Adam's report on this model.

# Misc

• The folder double-descent contains numerical experiments to observe a double descent phenomenon in the calibration curve on real datasets.