Conformal Prediction for Federated Uncertainty Quantification Under Label Shift
Authors: Vincent Plassier, Mehdi Makni, Aleksandr Rubashevskii, Eric Moulines, Maxim Panov
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental studies demonstrate that this method outperforms current competitors. and 5. Numerical experiments |
| Researcher Affiliation | Collaboration | 1Lagrange Mathematics and Computing Research Center, Paris, France 2CMAP, Ecole Polytechnique, Paris, France 3Skolkovo Institute of Science and Technology, Moscow, Russia 4Mohamed bin Zayed University of Artificial Intelligence, Masdar City, UAE 5Technology Innovation Institute, Abu Dhabi, UAE. |
| Pseudocode | Yes | Algorithm 1 DP-Fed Avg QE and Algorithm 2 DP-Fed CP |
| Open Source Code | No | The paper does not include an unambiguous statement or a direct link for the open-source code of the described methodology. |
| Open Datasets | Yes | CIFAR-10 Experiments. We investigate the performance of DP-Fed CP on the CIFAR-10 dataset. and Image Net Experiments. We use a pre-trained Res Net152 (He et al., 2016) as a base model with temperature scaling T = 10. We perform 1000 runs with different splits of the 50K Image Net test dataset into calibration and test datasets of size 40K and 10K samples, respectively. |
| Dataset Splits | Yes | We also randomly split the CIFAR-10 test dataset into a calibration dataset and a test dataset, each containing 5000 points, and repeat the experiment 1000 times. and We perform 1000 runs with different splits of the 50K Image Net test dataset into calibration and test datasets of size 40K and 10K samples, respectively. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, memory, or detailed computer specifications used for running its experiments. |
| Software Dependencies | No | The paper mentions models like ResNet but does not provide specific ancillary software details with version numbers (e.g., library or solver names with versions). |
| Experiment Setup | Yes | The optimization parameters taken for DP-Fed CP experiments are T = 200 iterations, γ = 1e-6 regularization parameter, η = 1e-3 stepsize, and K = 20 local iteration rounds. |