Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

Tighter CMI-Based Generalization Bounds via Stochastic Projection and Quantization

Authors: Milad Sefidgaran, Kimia Nadjahi, Abdellatif Zaidi

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical Our work is a theoretical paper with rigorously proven claims, and does not involve any experiment.
Researcher Affiliation Collaboration 1 Paris Research Center, Huawei Technologies France 2 CNRS, ENS Paris, France 3 Universit e Gustave Eiffel, France
Pseudocode No The considered noisy iterative optimization algorithm consists of the following steps: (Initialization) Sample Θ RD d and set the initial model s parameters to W0 = ΘW 0, where W 0 Rd. (Iterate) For t [T], apply the update rule W t = Proj n W t 1 ηt w b R(Vt, ΘW t 1) + σtεt o , (17) with ηt > 0 (the learning rate), σt 0 (the variance of the Gaussian noise), and εt N(0d, Id) (the isotropic Gaussian noise). Here, the projection is an optional operator often used to keep the norm of the model parameters bounded. (Output) Return the final hypothesis WT = ΘW T .
Open Source Code No Our work does not involve any experiment.
Open Datasets No Our work is a theoretical paper with rigorously proven claims, and does not involve any experiment.
Dataset Splits No Our work does not involve any experiment.
Hardware Specification No Our work does not involve any experiment.
Software Dependencies No Our work does not involve any experiment.
Experiment Setup No Our work does not involve any experiment.