meta-XB

This repository contains code for "Few-Shot Calibration of Set Predictors via Meta-Learned Cross-Validation-Based Conformal Prediction" - Sangwoo Park, Kfir M. Cohen, and Osvaldo Simeone.

Dependencies

This program is written in python 3.9 and uses PyTorch 1.10.2.

Essential Codes

• meta-XB can be found at meta_train/meta_training.py
• meta-VB can be found at meta_train/meta_tr_benchmark.py
• XB-CP can be found at funcs/jk_plus.py (for number of folds = number of examples) and funcs/cv_plus.py (for number of folds <= number of examples)
• VB-CP can be found at funcs/split_conformal.py
• soft inefficiency function is written in funcs/utils_for_set_prediction.py
• soft quantile via pinball loss (proposed way) and also via optimal transport (OT) can be found at funcs/soft_quantile.py
• further details can be found at the beginning of the main code main.py

Experiment on Multinomial Model and Inhomogeneous Features (Sec. V-A)

• runs_toy directory contains all the running shell script files

Experiment on Modulation Classification (Sec. V-B)

• runs_modulation_classification directory contains all the running shell script files

Experiment on miniImagenet Classification (Sec. V-C)

• runs_miniimagenet directory contains all the running shell script files

Experiment on Demodulation for Golden Angle Modulation (GAM) (Appendix C)

• runs_toy_vis_gam directory contains all the running shell script files