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..

A Gradient Guided Diffusion Framework for Chance Constrained Programming

Authors: Boyang Zhang, Zhiguo Wang, Ya-Feng Liu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on synthetic datasets and a waveform design task in wireless communications demonstrate that GGDOpt outperforms existing methods in both solution quality and stability with nearly 80% overhead reduction.
Researcher Affiliation Academia Boyang Zhang School of Advanced Interdisciplinary Sciences University of Chinese Academy of Sciences Beijing 100049, China EMAIL Zhiguo Wang Department of Mathematics Sichuan University Chengdu 610065, China EMAIL Ya-Feng Liu Ministry of Education Key Laboratory of Mathematics and Information Networks School of Mathematical Sciences Beijing University of Posts and Telecommunications Beijing 102206, China EMAIL
Pseudocode Yes Algorithm 1 Training of GGDOpt Input: {(x(i), ρ(i))}N i=1 p0(x|ρ). Output: sθ (x, t, ρ). 1: repeat 2: Load (x0, ρ0) p0(x|ρ). 3: Set ρ with probability puncond. 4: Sample t U[0, T]. 5: Generate xt p0t(xt|x0). 6: Take gradient descent step on (16). 7: until converged. Algorithm 2 Sampling of GGDOpt Input: sθ (x, t, ρ), objective f. Output: x 0. 1: x T p T . 2: for t = T, ..., 1 do 3: Calculate sθ(xt, t, ρ) with (18). 4: Calculate Gt with (11) or (12) . 5: Take guided sampling step with (17). 6: end for 7: return x 0 = x0.
Open Source Code Yes Our code is available at https://github.com/boyangzhang2000/GGDOpt.
Open Datasets No To generate training data, we solve the restricted version of problem (26) for N = 1000 values of z linearly spaced in the interval [0, 0.5]. Then we execute the reverse process with firstand second-order gradient guidance to generate samples.
Dataset Splits No The paper describes how training data is generated for synthetic problems and waveform design, but it does not specify explicit training/validation/test splits in terms of percentages, counts, or references to predefined splits for evaluating the GGDOpt model itself.
Hardware Specification No The paper states that hardware information is provided in the supplementary material but does not include specific hardware details (e.g., GPU/CPU models, memory) in the main text of the paper.
Software Dependencies No The paper mentions 'CVX (Grant et al. [2008])' but does not specify its version number. Other tools like 'Adam' and 'U-Net model' are mentioned but not with specific software versions.
Experiment Setup Yes In the diffusion and learning stage, we set T = 1000 with a linear noise schedule η(t) ranging from 0.0001 to 0.02, and let a(x, t) = 1/2η(t)x and b(t) = sqrt(η(t)). In the sampling with guidance stage, we evaluate both firstand second-order gradient guidances via implementing a DDIM-based technique (Song et al. [2020a]) with a descaled time step T = 100 for accelerated sampling. We employ two variants of the U-Net model (Ronneberger et al. [2015]) as our score estimator: U-Net-1D for the linear chance constrained problem and both for robust waveform design.