Jackpot: Approximating Uncertainty Domains with Adversarial Manifolds

Authors: Nathanaël Munier, Emmanuel Soubies, Pierre Weiss

JMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 3 Numerical Experiments 3.1 Measuring Masses in the Solar System 3.2 Blind Deblurring 3.3 Posterior Exploration for an Image Deblurring Problem
Researcher Affiliation Academia Nathana el Munier EMAIL University of Toulouse, CNRS, IMT, CBI, France Emmanuel Soubies University of Toulouse, CNRS, IRIT, CBI, France Pierre Weiss University of Toulouse, CNRS, IRIT, CBI, France
Pseudocode Yes Algorithm 1 BFS parameterization of Mε δ(Z)
Open Source Code Yes The numerical experiments supporting these results, together with the implementation of the Jackpot algorithm, are available in the corresponding Git Hub repository.
Open Datasets Yes Initial positions and speeds are taken from IMCCE (2019). URL https://ssp.imcce.fr/forms/ephemeris.
Dataset Splits No The paper discusses data acquisition parameters (e.g., 'measurements are taken on a period of 5 years within intervals of Δt = 7 days'), but does not specify training, validation, or test dataset splits in the context of model training or evaluation.
Hardware Specification No This work was performed using HPC resources from GENCI-IDRIS (Grant 2021-AD011012210R2).
Software Dependencies No It relies on the Python package deepinv (Tachella et al., 2025).
Experiment Setup No The paper mentions general strategies like 'using a gradient descent to high accuracy' and 'using the L-BFGS method', and provides some problem-specific parameters like 'N = 8' for Zernike coefficients and 'ε = 1000 km' for noise standard deviation. However, it lacks specific hyperparameter values for these optimization algorithms such as learning rates, batch sizes, or precise iteration counts.