Exploring-the-Noise-Robustness-of-Online-Conformal-Prediction

This repository is the official implementation of "Exploring the Noise Robustness of Online Conformal Prediction" at NIPS'2025.

Setup

1. Install uv (if not already installed):

curl -LsSf https://astral.sh/uv/install.sh | sh

2. Install dependencies:

uv sync

3. Configure data and model paths:

cp .env.example .env
# Edit .env and set paths for datasets and pre-trained models

Project Structure

src/
├── conformal.py       # Non-conformity scores (THR, APS, RAPS, SAPS) & threshold computation
├── utils.py           # Data loading, noise injection, and utility functions
├── config.py          # Environment configuration (.env)
├── models/            # Pre-trained models (ResNet, DenseNet, VGG)
└── saocp/             # SAOCP and rSAOCP implementations

Available Methods

Non-conformity Score Functions

• thr: Threshold (1 - softmax probability)
• aps: Adaptive Prediction Sets
• raps: Regularized Adaptive Prediction Sets
• saps: Sorted Adaptive Prediction Sets

Online Conformal Prediction Methods

• aci: Adaptive Conformal Inference (Gibbs & Candès, 2021)
• raci: Robust ACI (proposed, handles label noise)
• saocp: Strongly Adaptive Online Conformal Prediction (Bhatnagar et al., 2023)
• rsaocp: Robust SAOCP (proposed, handles label noise)

Citation

If you find this useful in your research, please consider citing:

@inproceedings{
    xi2025exploring,
    title={Exploring the Noise Robustness of Online Conformal Prediction},
    author={HuaJun Xi and Kangdao Liu and Hao Zeng and Wenguang Sun and Hongxin Wei},
    booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
    year={2025},
}