D-IMPL: A Diffusion-based Solver for Parameterized BBOs
Abstract
Diffusion models have demonstrated strong power in generative modeling tasks across multiple domains, exhibiting a remarkable capability of learning complex distributions from samples. In this paper, we leverage such capability to design an efficient universal diffusion-based solver for parameterized black-box optimizations (BBO), where the optimizer has only black-box access to queries of the objective function at the learning stage, yet is able to reduce the additional computational cost at the inference stage for each BBO instance while also capturing the potential multi-modal landscape of non-convex objectives. To cast our formulation as a compatible generative modeling task, we introduce the notion of minimization policy as a new solution concept, which defines a sampling distribution over the solutions that should concentrate around the minimizer set for each BBO instance. We then propose Diffusion-based Iterative Minimization Policy Learning (D-IMPL), a practical generative-model-based solver for solving parameterized BBOs that employs diffusion models to learn a minimization policy, whose density is proportional to the exponential of the negated objective values, thereby amortizing the computational costs across different BBO instances. Furthermore, we demonstrate the performance of our D-IMPL algorithm by establishing a sample complexity guarantee showing that a δ-approximate minimization policy can be effectively learned within O(log(1/δ)) iterations, and by extensive empirical evaluations over a range of constrained and unconstrained BBO tasks.
Related
- Height functions on singular surfaces parameterized by smooth maps 𝒜-equivalent to Hk
- Parameterized partial theta identities and a unified q-difference proof
- A parameterized block LU preconditioning for double saddle point problems
- A Gaussian Process Regression-Enhanced Two-Sweep Parameterized Power Method for Computing Extreme Eigenpairs
- Inertial parameterized Douglas-Rachford for generalized DC programming problems
BibTeX
@misc{hu2026dimpl,
title = {{D-IMPL: A Diffusion-based Solver for Parameterized BBOs}},
author = {Yang Hu and Na Li},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.22752},
eprint = {2609.22752},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.22752},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - D-IMPL: A Diffusion-based Solver for Parameterized BBOs
AU - Yang Hu
AU - Na Li
PY - 2026
DA - 2026/09/19
PB - arXiv
DO - 10.48550/arxiv.2609.22752
UR - https://arxiv.org/abs/2609.22752
AB - Diffusion models have demonstrated strong power in generative modeling tasks across multiple domains, exhibiting a remarkable capability of learning complex distributions from samples. In this paper, we leverage such capability to design an efficient universal diffusion-based solver for parameterized black-box optimizations (BBO), where the optimizer has only black-box access to queries of the objective function at the learning stage, yet is able to reduce the additional computational cost at the inference stage for each BBO instance while also capturing the potential multi-modal landscape of non-convex objectives. To cast our formulation as a compatible generative modeling task, we introduce the notion of minimization policy as a new solution concept, which defines a sampling distribution over the solutions that should concentrate around the minimizer set for each BBO instance. We then propose Diffusion-based Iterative Minimization Policy Learning (D-IMPL), a practical generative-model-based solver for solving parameterized BBOs that employs diffusion models to learn a minimization policy, whose density is proportional to the exponential of the negated objective values, thereby amortizing the computational costs across different BBO instances. Furthermore, we demonstrate the performance of our D-IMPL algorithm by establishing a sample complexity guarantee showing that a δ-approximate minimization policy can be effectively learned within O(log(1/δ)) iterations, and by extensive empirical evaluations over a range of constrained and unconstrained BBO tasks.
ER -