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..

Towards Understanding the Mechanisms of Classifier-Free Guidance

Authors: Xiang Li, Rongrong Wang, Qing Qu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We then verify these insights in real-world, nonlinear diffusion models: over a broad range of noise levels, linear CFG resembles the behavior of its nonlinear counterpart. Although the two eventually diverge at low noise levels, we discuss how the insights from the linear analysis still shed light on the CFG s mechanism in the nonlinear regime.
Researcher Affiliation Academia 1Department of EECS, University of Michigan, 2Department of CMSE and Mathematics, Michigan State University, EMAIL, EMAIL, EMAIL
Pseudocode No The paper includes mathematical equations, derivations, and descriptions of methods, but it does not contain any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured steps in a code-like format.
Open Source Code No The code and instructions for reproducing our experiments will be released at: https://github. com/Morefre/Towards-Understanding-the-Mechanisms-of-Classifier-Free-Guidance. git.
Open Datasets Yes We perform our experiments on CIFAR-10 [42] and Image Net dataset [29]
Dataset Splits No The paper does not explicitly provide training/test/validation dataset splits with percentages, counts, or predefined citations. It mentions using the CIFAR-10 and ImageNet datasets and generating samples from a trained diffusion model, but it does not detail how these datasets were originally split for training or evaluation.
Hardware Specification Yes All experiments are performed on A100 GPUs with 80 GB memory.
Software Dependencies No The paper mentions using specific models like EDM-1 and EDM-2, but does not specify version numbers for programming languages (e.g., Python) or libraries (e.g., PyTorch, TensorFlow, CUDA).
Experiment Setup Yes CFG is applied to the entire noise interval σ(t) [0.002, 80], with guidance strength γ = 4. The samples are generated using 20 steps of Euler method (first-order sampler).