Input uncertainty propagation through trained neural networks

Authors: Paul Monchot, Loic Coquelin, Sébastien Julien Petit, Sébastien Marmin, Erwan Le Pennec, Nicolas Fischer

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The methodology is tested against a wide range of datasets and networks. It shows robustness, and genericity and offers highly accurate output probability density function estimation while maintaining a reasonable computational cost compared with the standard Monte Carlo (MC) approach.
Researcher Affiliation Academia 1CMAP, CNRS, Ecole polytechnique, Institut Polytechnique de Paris, Palaiseau, France 2Data Science and Uncertainty Department, National Laboratory of Metrology and Testing, Paris, France.
Pseudocode Yes Algorithm 1 presents the full algorithm.
Open Source Code Yes 1The code to reproduce the experiments is available at https: //github.com/Paul Mcht/WGMprop.
Open Datasets Yes Finally, extensive experimental results demonstrate that the proposed method performs favorably against the state-of-the art approaches on the MNIST (Le Cun, 1998), CIFAR10 (Krizhevsky et al., 2009), CIFAR100 (Krizhevsky et al., 2009), and Camelyon (Litjens et al., 2018) datasets.
Dataset Splits No The paper mentions using 'test images' (e.g., '500 test images from the MNIST dataset') and standard datasets like MNIST and CIFAR, which commonly have predefined splits. However, it does not explicitly state the specific training, validation, and test dataset splits (e.g., percentages, counts, or a reference to specific split files) used for reproducibility of the data partitioning.
Hardware Specification Yes The experiments were conducted using a GPU NVIDIA Telsa V100 32GO HBM2 and a CPU Intel Xeon Silver 4210.
Software Dependencies Yes All the codes were written using Python 3 (Van Rossum & Drake, 2009) and Tensorflow 2 (Agrawal et al., 2019).
Experiment Setup Yes Hyperparameters: In the following of this section experimental results are displayed using the following parameters: s0 = 0, Tsplit = 0.0001, no subspace selection (r = input dimension).